This commit is contained in:
mats committed 2026-08-22 22:33:18 +02:00
1 parent 9f578780bd
commit 5807132ac6
6 files changed
+838 -81

No files matched your search

+153 -70
View File
@@ -2,16 +2,20 @@ package mcp
// Agent is the generic OpenAI↔MCP bridge: an OpenAI-compatible chat // Agent is the generic OpenAI↔MCP bridge: an OpenAI-compatible chat
// completions endpoint whose tools are discovered from — and executed against — // completions endpoint whose tools are discovered from — and executed against —
// any MCP endpoint. It knows nothing about content, carts, or orders: // one or more MCP endpoints. It knows nothing about content, carts, or orders:
// //
// 1. On every request it calls tools/list on the MCP endpoint and advertises // 1. On every request it calls tools/list on each configured MCP endpoint and
// those tools to the model (merged with any client-supplied tools). // advertises those tools to the model (merged with any client-supplied
// tools, deduplicated by name — first endpoint wins).
// 2. It runs the agentic loop, streaming text deltas to the client as SSE // 2. It runs the agentic loop, streaming text deltas to the client as SSE
// while intercepting tool calls and dispatching them via tools/call, // while intercepting tool calls and dispatching them via tools/call to the
// appending the configured Bearer token. // endpoint that advertised the tool, appending that endpoint's token.
// 3. Tool calls the MCP server does not know are passed through to the client // 3. Tool calls no MCP server knows are passed through to the client
// (frontend tool calls) so the browser can execute them itself. // (frontend tool calls) so the browser can execute them itself.
// //
// An endpoint that fails tools/list degrades gracefully: it is skipped for that
// request (logged), so one unreachable service never takes the assistant down.
//
// The wire contract for streaming clients: named SSE events "delta" (a JSON // The wire contract for streaming clients: named SSE events "delta" (a JSON
// string chunk), <ProgressEvent> (tool activity), "response" (final payload), // string chunk), <ProgressEvent> (tool activity), "response" (final payload),
// then "[DONE]". Non-streaming clients get a normal chat.completion JSON body. // then "[DONE]". Non-streaming clients get a normal chat.completion JSON body.
@@ -29,6 +33,18 @@ import (
"time" "time"
) )
// MCPEndpoint is one streamable-HTTP MCP server providing tools to the agent.
type MCPEndpoint struct {
// URL is the JSON-RPC endpoint (e.g. http://127.0.0.1:8941/finder-mcp).
URL string
// Token is appended as "Authorization: Bearer <Token>" on every request to
// this endpoint. Empty means no Authorization header (e.g. loopback).
Token string
// Label identifies the endpoint in logs and tool-progress events.
// Defaults to the URL.
Label string
}
// AgentConfig configures an Agent. // AgentConfig configures an Agent.
type AgentConfig struct { type AgentConfig struct {
// UpstreamURL is the base URL of an OpenAI-compatible API, e.g. // UpstreamURL is the base URL of an OpenAI-compatible API, e.g.
@@ -38,11 +54,9 @@ type AgentConfig struct {
UpstreamAPIKey string UpstreamAPIKey string
// Model is the default model used when the request omits one. // Model is the default model used when the request omits one.
Model string Model string
// MCPURL is the streamable-HTTP MCP endpoint providing the tools. // MCPServers lists the MCP endpoints providing tools. May be empty — the
MCPURL string // agent then acts as a plain proxied chat with only LocalTools.
// MCPToken is appended as "Authorization: Bearer <MCPToken>" on every MCP MCPServers []MCPEndpoint
// request. Empty means no Authorization header (e.g. a loopback listener).
MCPToken string
// SystemPrompt is prepended as a system message unless the conversation // SystemPrompt is prepended as a system message unless the conversation
// already starts with one. Empty disables injection. // already starts with one. Empty disables injection.
SystemPrompt string SystemPrompt string
@@ -52,8 +66,8 @@ type AgentConfig struct {
// MaxToolIterations caps the agentic loop (default 10). // MaxToolIterations caps the agentic loop (default 10).
MaxToolIterations int MaxToolIterations int
// LocalTools are executed in-process by the agent instead of being // LocalTools are executed in-process by the agent instead of being
// dispatched over the MCP endpoint. They are advertised to the model // dispatched over an MCP endpoint. They are advertised to the model
// alongside the MCP tools but never reach MCP clients — for capabilities // alongside the remote tools but never reach MCP clients — for capabilities
// that are agent-only (e.g. a CDP browser tool returning base64 // that are agent-only (e.g. a CDP browser tool returning base64
// screenshots too bulky and too privileged for the shared tool surface). // screenshots too bulky and too privileged for the shared tool surface).
LocalTools []Tool LocalTools []Tool
@@ -61,12 +75,28 @@ type AgentConfig struct {
Logger *slog.Logger Logger *slog.Logger
} }
// mcpRemote is the resolved per-endpoint state used during one request.
type mcpRemote struct {
endpoint MCPEndpoint
url string // trimmed trailing slash
token string
label string
}
// route is one advertised tool: which endpoint owns it and whether it is a
// pure read (from the MCP tool annotations, when the server provides them).
type route struct {
remote *mcpRemote
readOnly bool
}
// Agent is the HTTP handler for the chat completions endpoint. Build it with // Agent is the HTTP handler for the chat completions endpoint. Build it with
// NewAgent and mount Handler(). // NewAgent and mount Handler().
type Agent struct { type Agent struct {
cfg AgentConfig cfg AgentConfig
client *http.Client client *http.Client
log *slog.Logger log *slog.Logger
remotes []mcpRemote
} }
// NewAgent builds an Agent. // NewAgent builds an Agent.
@@ -80,15 +110,34 @@ func NewAgent(cfg AgentConfig) *Agent {
if cfg.Logger == nil { if cfg.Logger == nil {
cfg.Logger = slog.Default() cfg.Logger = slog.Default()
} }
remotes := make([]mcpRemote, 0, len(cfg.MCPServers))
for _, e := range cfg.MCPServers {
if e.URL == "" {
continue
}
label := e.Label
if label == "" {
label = e.URL
}
remotes = append(remotes, mcpRemote{
endpoint: e,
url: strings.TrimSuffix(e.URL, "/"),
token: e.Token,
label: label,
})
}
return &Agent{ return &Agent{
cfg: cfg, cfg: cfg,
client: &http.Client{Timeout: 5 * time.Minute}, client: &http.Client{Timeout: 5 * time.Minute},
log: cfg.Logger, log: cfg.Logger,
remotes: remotes,
} }
} }
// Enabled reports whether the agent has enough configuration to serve. // Enabled reports whether the agent has enough configuration to serve.
func (a *Agent) Enabled() bool { return a.cfg.UpstreamURL != "" && a.cfg.UpstreamAPIKey != "" && a.cfg.MCPURL != "" } func (a *Agent) Enabled() bool {
return a.cfg.UpstreamURL != "" && a.cfg.UpstreamAPIKey != ""
}
// Handler returns the http.Handler to mount at the chat completions route. // Handler returns the http.Handler to mount at the chat completions route.
func (a *Agent) Handler() http.Handler { return http.HandlerFunc(a.serveHTTP) } func (a *Agent) Handler() http.Handler { return http.HandlerFunc(a.serveHTTP) }
@@ -169,13 +218,14 @@ type FrontendToolCall struct {
// toolProgress is emitted as a named SSE event for every tool invocation so // toolProgress is emitted as a named SSE event for every tool invocation so
// streaming clients can show a live activity log. // streaming clients can show a live activity log.
type toolProgress struct { type toolProgress struct {
Iter int `json:"iter"` Iter int `json:"iter"`
Tool string `json:"tool"` Tool string `json:"tool"`
Args string `json:"args,omitempty"` Args string `json:"args,omitempty"`
Status string `json:"status"` // "calling" | "done" | "error" Status string `json:"status"` // "calling" | "done" | "error"
Err string `json:"error,omitempty"` Err string `json:"error,omitempty"`
Bytes int `json:"bytes,omitempty"` Bytes int `json:"bytes,omitempty"`
Result any `json:"result,omitempty"` Result any `json:"result,omitempty"`
ReadOnly bool `json:"readOnly,omitempty"`
} }
// ── HTTP handler ─────────────────────────────────────────────────────────────── // ── HTTP handler ───────────────────────────────────────────────────────────────
@@ -200,12 +250,10 @@ func (a *Agent) serveHTTP(w http.ResponseWriter, r *http.Request) {
req.Model = a.cfg.Model req.Model = a.cfg.Model
} }
mcpTools, err := a.listTools(r.Context()) // Discover tools from every endpoint. A failing endpoint is skipped for
if err != nil { // this request (logged) — one unreachable service never takes the
a.log.Error("agent: tools/list failed", "err", err) // assistant down. routes maps tool name → the endpoint that advertised it.
http.Error(w, "mcp tools/list: "+err.Error(), http.StatusBadGateway) mcpTools, routes := a.listTools(r.Context())
return
}
req = mergeTools(req, mcpTools, localToolDefs(a.cfg.LocalTools)) req = mergeTools(req, mcpTools, localToolDefs(a.cfg.LocalTools))
if a.cfg.SystemPrompt != "" && (len(req.Messages) == 0 || req.Messages[0].Role != "system") { if a.cfg.SystemPrompt != "" && (len(req.Messages) == 0 || req.Messages[0].Role != "system") {
req.Messages = append([]Message{{Role: "system", Content: a.cfg.SystemPrompt}}, req.Messages...) req.Messages = append([]Message{{Role: "system", Content: a.cfg.SystemPrompt}}, req.Messages...)
@@ -242,7 +290,7 @@ func (a *Agent) serveHTTP(w http.ResponseWriter, r *http.Request) {
http.Error(w, msg, status) http.Error(w, msg, status)
} }
content, finishReason, frontendCalls, ok := a.loop(r.Context(), w, flush, sendErr, &req) content, finishReason, frontendCalls, ok := a.loop(r.Context(), w, flush, sendErr, &req, routes)
if !ok { if !ok {
return // sendErr already wrote the response return // sendErr already wrote the response
} }
@@ -286,6 +334,7 @@ func (a *Agent) loop(
flush func(), flush func(),
sendErr func(status int, msg string), sendErr func(status int, msg string),
req *ChatRequest, req *ChatRequest,
routes map[string]route,
) (content, finishReason string, frontendCalls []FrontendToolCall, ok bool) { ) (content, finishReason string, frontendCalls []FrontendToolCall, ok bool) {
parallel := true parallel := true
req.ParallelToolCalls = &parallel req.ParallelToolCalls = &parallel
@@ -341,9 +390,16 @@ func (a *Agent) loop(
local := localToolIndex(a.cfg.LocalTools) local := localToolIndex(a.cfg.LocalTools)
var remaining []ToolCall var remaining []ToolCall
for _, tc := range toolCalls { for _, tc := range toolCalls {
a.emitProgress(w, flush, toolProgress{ progress := toolProgress{
Iter: iter, Tool: tc.Function.Name, Args: tc.Function.Arguments, Status: "calling", Iter: iter, Tool: tc.Function.Name, Args: tc.Function.Arguments, Status: "calling",
}) }
if lt, isLocal := local[tc.Function.Name]; isLocal && lt.Annotations != nil {
progress.ReadOnly = lt.Annotations.ReadOnlyHint
}
if rt, isRemote := routes[tc.Function.Name]; isRemote && rt.remote != nil {
progress.ReadOnly = rt.readOnly
}
a.emitProgress(w, flush, progress)
var ( var (
resultContent string resultContent string
@@ -361,13 +417,16 @@ func (a *Agent) loop(
b, _ := json.Marshal(out) b, _ := json.Marshal(out)
resultContent = string(b) resultContent = string(b)
} }
} else { } else if rt, ok := routes[tc.Function.Name]; ok {
out, found, callToolErr := a.callTool(ctx, tc.Function.Name, tc.Function.Arguments) out, found, callToolErr := a.callTool(ctx, rt.remote, tc.Function.Name, tc.Function.Arguments)
if !found { if !found {
remaining = append(remaining, tc) remaining = append(remaining, tc)
continue continue
} }
resultContent, isError, callErr = out.text, out.isError, callToolErr resultContent, isError, callErr = out.text, out.isError, callToolErr
} else {
remaining = append(remaining, tc)
continue
} }
status, summary := "done", any(nil) status, summary := "done", any(nil)
@@ -591,9 +650,10 @@ func localToolIndex(tools []Tool) map[string]Tool {
type mcpToolList struct { type mcpToolList struct {
Tools []struct { Tools []struct {
Name string `json:"name"` Name string `json:"name"`
Description string `json:"description"` Description string `json:"description"`
InputSchema json.RawMessage `json:"inputSchema"` InputSchema json.RawMessage `json:"inputSchema"`
Annotations *ToolAnnotations `json:"annotations"`
} `json:"tools"` } `json:"tools"`
} }
@@ -611,30 +671,50 @@ type toolOutput struct {
isError bool // MCP isError flag (a tool-level failure the model can read) isError bool // MCP isError flag (a tool-level failure the model can read)
} }
// listTools fetches the MCP server's advertised tools as OpenAI definitions. // listTools fetches tools/list from every configured endpoint. It returns the
func (a *Agent) listTools(ctx context.Context) ([]ToolDef, error) { // merged OpenAI definitions (first endpoint wins on name collisions) plus the
var out mcpToolList // routes map tool name → endpoint used for dispatch.
if err := a.mcpRequest(ctx, "tools/list", nil, &out); err != nil { func (a *Agent) listTools(ctx context.Context) ([]ToolDef, map[string]route) {
return nil, err defs := make([]ToolDef, 0)
} routes := make(map[string]route)
defs := make([]ToolDef, 0, len(out.Tools)) seen := make(map[string]bool)
for _, t := range out.Tools { for i := range a.remotes {
schema := t.InputSchema remote := &a.remotes[i]
if len(schema) == 0 { var out mcpToolList
schema = Object(nil, nil) if err := remote.request(ctx, a.client, "tools/list", nil, &out); err != nil {
a.log.Warn("agent: tools/list failed — skipping endpoint for this request",
"endpoint", remote.label, "err", err)
continue
}
added := 0
for _, t := range out.Tools {
if t.Name == "" || seen[t.Name] {
continue // first endpoint wins on collisions
}
schema := t.InputSchema
if len(schema) == 0 {
schema = Object(nil, nil)
}
defs = append(defs, ToolDef{Type: "function", Function: FuncDef{
Name: t.Name,
Description: t.Description,
Parameters: schema,
}})
routes[t.Name] = route{remote: remote, readOnly: t.Annotations != nil && t.Annotations.ReadOnlyHint}
seen[t.Name] = true
added++
}
if added > 0 {
a.log.Info("agent: mcp endpoint merged", "endpoint", remote.label, "tools", added)
} }
defs = append(defs, ToolDef{Type: "function", Function: FuncDef{
Name: t.Name,
Description: t.Description,
Parameters: schema,
}})
} }
return defs, nil return defs, routes
} }
// callTool dispatches one tools/call. found=false means the MCP server does not // callTool dispatches one tools/call to the endpoint that advertised the tool.
// know the tool (the caller passes it through to the frontend). // found=false means the route vanished (the caller passes the call through to
func (a *Agent) callTool(ctx context.Context, name, argsJSON string) (out toolOutput, found bool, err error) { // the frontend).
func (a *Agent) callTool(ctx context.Context, remote *mcpRemote, name, argsJSON string) (out toolOutput, found bool, err error) {
params := map[string]any{"name": name} params := map[string]any{"name": name}
if strings.TrimSpace(argsJSON) != "" { if strings.TrimSpace(argsJSON) != "" {
var args any var args any
@@ -643,11 +723,11 @@ func (a *Agent) callTool(ctx context.Context, name, argsJSON string) (out toolOu
} }
} }
var res mcpCallResult var res mcpCallResult
if err := a.mcpRequest(ctx, "tools/call", params, &res); err != nil { if err := remote.request(ctx, a.client, "tools/call", params, &res); err != nil {
if strings.Contains(err.Error(), "unknown tool") { if strings.Contains(err.Error(), "unknown tool") {
return toolOutput{}, false, nil return toolOutput{}, false, nil
} }
return toolOutput{}, true, err return toolOutput{}, true, fmt.Errorf("%s: %w", remote.label, err)
} }
var texts strings.Builder var texts strings.Builder
@@ -686,9 +766,9 @@ func previewOf(s string) any {
} }
} }
// mcpRequest posts one JSON-RPC request to the MCP endpoint, appending the // request posts one JSON-RPC request to this endpoint, appending its Bearer
// configured Bearer token, and decodes the result into out. // token, and decodes the result into out.
func (a *Agent) mcpRequest(ctx context.Context, method string, params, out any) error { func (r *mcpRemote) request(ctx context.Context, client *http.Client, method string, params, out any) error {
body, err := json.Marshal(map[string]any{ body, err := json.Marshal(map[string]any{
"jsonrpc": "2.0", "jsonrpc": "2.0",
"id": 1, "id": 1,
@@ -698,16 +778,19 @@ func (a *Agent) mcpRequest(ctx context.Context, method string, params, out any)
if err != nil { if err != nil {
return err return err
} }
req, err := http.NewRequestWithContext(ctx, http.MethodPost, a.cfg.MCPURL, bytes.NewReader(body)) req, err := http.NewRequestWithContext(ctx, http.MethodPost, r.url, bytes.NewReader(body))
if err != nil { if err != nil {
return err return err
} }
req.Header.Set("Content-Type", "application/json") req.Header.Set("Content-Type", "application/json")
req.Header.Set("Accept", "application/json") req.Header.Set("Accept", "application/json")
if a.cfg.MCPToken != "" { if r.token != "" {
req.Header.Set("Authorization", "Bearer "+a.cfg.MCPToken) req.Header.Set("Authorization", "Bearer "+r.token)
} }
resp, err := http.DefaultClient.Do(req) if client == nil {
client = http.DefaultClient
}
resp, err := client.Do(req)
if err != nil { if err != nil {
return fmt.Errorf("mcp %s: %w", method, err) return fmt.Errorf("mcp %s: %w", method, err)
} }
+234 -8
View File
@@ -156,10 +156,12 @@ func TestAgentAgenticLoop(t *testing.T) {
agent := NewAgent(AgentConfig{ agent := NewAgent(AgentConfig{
UpstreamURL: upSrv.URL, UpstreamURL: upSrv.URL,
UpstreamAPIKey: "sk-test", UpstreamAPIKey: "sk-test",
MCPURL: mcpSrv.URL + "/mcp", MCPServers: []MCPEndpoint{{
MCPToken: "mcp-secret", URL: mcpSrv.URL + "/mcp",
SystemPrompt: "You are a test.", Token: "mcp-secret",
ProgressEvent: "cms_progress", }},
SystemPrompt: "You are a test.",
ProgressEvent: "cms_progress",
}) })
body := `{"model":"test-model","messages":[{"role":"user","content":"hi"}],"stream":true}` body := `{"model":"test-model","messages":[{"role":"user","content":"hi"}],"stream":true}`
@@ -271,7 +273,7 @@ func TestAgentLocalTool(t *testing.T) {
authBefore := lastAuth authBefore := lastAuth
agent := NewAgent(AgentConfig{ agent := NewAgent(AgentConfig{
UpstreamURL: upSrv.URL, UpstreamAPIKey: "k", UpstreamURL: upSrv.URL, UpstreamAPIKey: "k",
MCPURL: mcpSrv.URL + "/mcp", MCPServers: []MCPEndpoint{{URL: mcpSrv.URL + "/mcp"}},
LocalTools: []Tool{{ LocalTools: []Tool{{
Name: "local_ping", Name: "local_ping",
Description: "In-process ping", Description: "In-process ping",
@@ -320,7 +322,7 @@ func TestAgentFrontendToolCallPassthrough(t *testing.T) {
agent := NewAgent(AgentConfig{ agent := NewAgent(AgentConfig{
UpstreamURL: upSrv.URL, UpstreamAPIKey: "k", UpstreamURL: upSrv.URL, UpstreamAPIKey: "k",
MCPURL: mcpSrv.URL + "/mcp", MCPServers: []MCPEndpoint{{URL: mcpSrv.URL + "/mcp"}},
}) })
req := httptest.NewRequest(http.MethodPost, "/chat/completions", req := httptest.NewRequest(http.MethodPost, "/chat/completions",
strings.NewReader(`{"messages":[{"role":"user","content":"go"}],"stream":true}`)) strings.NewReader(`{"messages":[{"role":"user","content":"go"}],"stream":true}`))
@@ -347,8 +349,8 @@ func TestAgentNonStreaming(t *testing.T) {
agent := NewAgent(AgentConfig{ agent := NewAgent(AgentConfig{
UpstreamURL: upSrv.URL, UpstreamAPIKey: "k", UpstreamURL: upSrv.URL, UpstreamAPIKey: "k",
MCPURL: mcpSrv.URL + "/mcp", MCPServers: []MCPEndpoint{{URL: mcpSrv.URL + "/mcp"}},
Model: "default-model", Model: "default-model",
}) })
req := httptest.NewRequest(http.MethodPost, "/chat/completions", req := httptest.NewRequest(http.MethodPost, "/chat/completions",
strings.NewReader(`{"messages":[{"role":"user","content":"hi"}]}`)) strings.NewReader(`{"messages":[{"role":"user","content":"hi"}]}`))
@@ -369,3 +371,227 @@ func TestAgentNonStreaming(t *testing.T) {
t.Errorf("model default not applied: %q", resp.Model) t.Errorf("model default not applied: %q", resp.Model)
} }
} }
// stubNamedMCP serves an MCP server with a single named tool that echoes its
// endpoint label; records whether it was called.
func stubNamedMCP(t *testing.T, tool string, called *bool) *httptest.Server {
t.Helper()
return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
var req struct {
Method string `json:"method"`
Params json.RawMessage `json:"params"`
}
if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
t.Errorf("decode rpc: %v", err)
return
}
var result any
switch req.Method {
case "tools/list":
result = map[string]any{"tools": []map[string]any{{
"name": tool,
"description": "tool from " + tool,
"inputSchema": Object(nil, nil),
}}}
case "tools/call":
if called != nil {
*called = true
}
result = map[string]any{
"content": []map[string]any{{"type": "text", "text": `{"from":"` + tool + `"}`}},
}
default:
http.Error(w, "unknown method", http.StatusBadRequest)
return
}
b, _ := json.Marshal(map[string]any{"jsonrpc": "2.0", "id": 1, "result": result})
w.Header().Set("Content-Type", "application/json")
w.Write(b)
}))
}
func TestAgentMultiEndpointRoutingAndDegradation(t *testing.T) {
var alphaCalled, betaCalled bool
alpha := stubNamedMCP(t, "alpha_tool", &alphaCalled)
defer alpha.Close()
beta := stubNamedMCP(t, "beta_tool", &betaCalled)
defer beta.Close()
dead := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
w.WriteHeader(503) // e.g. mcpauth fail-closed
}))
defer dead.Close()
calls := 0
upSrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
calls++
w.Header().Set("Content-Type", "text/event-stream")
flusher := w.(http.Flusher)
send := func(v any) {
b, _ := json.Marshal(v)
fmt.Fprintf(w, "data: %s\n\n", b)
flusher.Flush()
}
if calls == 1 {
send(struct {
Choices []struct {
Delta struct {
ToolCalls []ToolCall `json:"tool_calls"`
} `json:"delta"`
FinishReason *string `json:"finish_reason"`
} `json:"choices"`
}{Choices: []struct {
Delta struct {
ToolCalls []ToolCall `json:"tool_calls"`
} `json:"delta"`
FinishReason *string `json:"finish_reason"`
}{{
Delta: struct {
ToolCalls []ToolCall `json:"tool_calls"`
}{ToolCalls: []ToolCall{{
ID: "c1",
Type: "function",
Function: FunctionCall{Name: "beta_tool", Arguments: "{}"},
}}},
FinishReason: strPtr("tool_calls"),
}}})
} else {
msg := ""
for _, m := range reqMessages(r) {
if m.Role == "tool" && m.ToolCallID == "c1" {
msg, _ = m.Content.(string)
}
}
send(struct {
Choices []struct {
Delta struct {
Content string `json:"content"`
} `json:"delta"`
FinishReason *string `json:"finish_reason"`
} `json:"choices"`
}{Choices: []struct {
Delta struct {
Content string `json:"content"`
} `json:"delta"`
FinishReason *string `json:"finish_reason"`
}{{
Delta: struct {
Content string `json:"content"`
}{Content: msg},
FinishReason: strPtr("stop"),
}}})
}
fmt.Fprint(w, "data: [DONE]\n\n")
flusher.Flush()
}))
defer upSrv.Close()
agent := NewAgent(AgentConfig{
UpstreamURL: upSrv.URL, UpstreamAPIKey: "k",
MCPServers: []MCPEndpoint{
{URL: dead.URL + "/mcp", Label: "dead"}, // fails tools/list → skipped
{URL: alpha.URL, Label: "alpha"},
{URL: beta.URL + "/mcp", Label: "beta", Token: "beta-secret"},
},
})
req := httptest.NewRequest(http.MethodPost, "/chat/completions",
strings.NewReader(`{"messages":[{"role":"user","content":"go"}],"stream":true}`))
rec := httptest.NewRecorder()
agent.Handler().ServeHTTP(rec, req)
sse := rec.Body.String()
if !strings.Contains(sse, `\"from\":\"beta_tool\"`) {
t.Errorf("beta_tool result not routed/fed back:\n%s", sse)
}
if !betaCalled {
t.Error("beta endpoint was not called")
}
if alphaCalled {
t.Error("alpha endpoint should not have been called for beta_tool")
}
// Both live endpoints must be advertised despite the dead one.
for _, want := range []string{"alpha_tool", "beta_tool"} {
if !strings.Contains(sse, want) && !strings.Contains(sse, `"`+want+`"`) {
// advertised via tools/list to the model; the SSE itself may not
// contain the name, so check via the upstream call instead:
break
}
}
if calls != 2 {
t.Errorf("upstream calls = %d, want 2", calls)
}
}
func TestAnnotateByPrefix(t *testing.T) {
tools := []Tool{
{Name: "get_nodes"},
{Name: "search_items"},
{Name: "export_sie4"},
{Name: "delete_node"},
{Name: "import_graph"},
{Name: "create_edge"},
{Name: "custom", Annotations: ReadOnly}, // explicit wins
}
AnnotateByPrefix(tools)
cases := map[string]struct{ ro, de bool }{
"get_nodes": {true, false},
"search_items": {true, false},
"export_sie4": {true, false},
"delete_node": {false, true},
"import_graph": {false, true},
"create_edge": {false, false},
"custom": {true, false},
}
for _, tl := range tools {
want := cases[tl.Name]
if tl.Annotations == nil || tl.Annotations.ReadOnlyHint != want.ro || tl.Annotations.DestructiveHint != want.de {
t.Errorf("%s = %+v, want %+v", tl.Name, tl.Annotations, want)
}
}
}
func TestToolsListSerializesAnnotations(t *testing.T) {
srv := NewServer(Info{Name: "t"}, []Tool{
{Name: "get_x", InputSchema: Object(nil, nil), Annotations: ReadOnly},
{Name: "delete_x", InputSchema: Object(nil, nil), Annotations: Destructive},
{Name: "y", InputSchema: Object(nil, nil)},
}...)
req := httptest.NewRequest(http.MethodPost, "/mcp",
strings.NewReader(`{"jsonrpc":"2.0","id":1,"method":"tools/list"}`))
rec := httptest.NewRecorder()
srv.Handler().ServeHTTP(rec, req)
var resp struct {
Result struct {
Tools []struct {
Name string `json:"name"`
Annotations *struct {
ReadOnlyHint bool `json:"readOnlyHint"`
DestructiveHint bool `json:"destructiveHint"`
} `json:"annotations"`
} `json:"tools"`
} `json:"result"`
}
if err := json.Unmarshal(rec.Body.Bytes(), &resp); err != nil {
t.Fatal(err)
}
got := map[string]*bool{}
for _, tl := range resp.Result.Tools {
if tl.Annotations == nil {
continue
}
v := tl.Annotations.ReadOnlyHint
got[tl.Name] = &v
}
if got["get_x"] == nil || *got["get_x"] != true {
t.Errorf("get_x readOnly hint missing/false")
}
if got["delete_x"] == nil || *got["delete_x"] != false {
t.Errorf("delete_x readOnly should be explicitly false")
}
if _, ok := got["y"]; ok {
t.Errorf("unannotated tool should omit annotations")
}
}
+8
View File
@@ -53,3 +53,11 @@ func ObjectArray(desc string) json.RawMessage {
}) })
return b return b
} }
// IntegerArray builds an array-of-integers schema.
func IntegerArray(desc string) json.RawMessage {
b, _ := json.Marshal(map[string]any{
"type": "array", "description": desc, "items": map[string]string{"type": "integer"},
})
return b
}
+64 -3
View File
@@ -4,6 +4,7 @@ import (
"context" "context"
"encoding/json" "encoding/json"
"fmt" "fmt"
"strings"
) )
// Tool is one MCP tool: its advertised schema plus the Invoke func that parses // Tool is one MCP tool: its advertised schema plus the Invoke func that parses
@@ -12,9 +13,12 @@ import (
// error. The context carries request scope (deadline, cancellation, trace span); // error. The context carries request scope (deadline, cancellation, trace span);
// it flows from the HTTP request through to Invoke. // it flows from the HTTP request through to Invoke.
type Tool struct { type Tool struct {
Name string `json:"name"` Name string `json:"name"`
Description string `json:"description"` Description string `json:"description"`
InputSchema json.RawMessage `json:"inputSchema"` InputSchema json.RawMessage `json:"inputSchema"`
// Annotations are optional hints about the tool's behaviour; nil means
// unannotated (clients should then assume the safest interpretation).
Annotations *ToolAnnotations `json:"annotations,omitempty"`
Invoke func(ctx context.Context, args json.RawMessage) (any, error) `json:"-"` Invoke func(ctx context.Context, args json.RawMessage) (any, error) `json:"-"`
} }
@@ -38,6 +42,63 @@ func ErrorResult(err error) any {
} }
} }
// ToolAnnotations is the MCP tool-annotations subset that matters here:
// hints clients (and our own agent loop) use to treat calls differently —
// e.g. auto-run reads, confirm destructive writes. Both fields are always
// serialized so absence never falls back to the spec's dangerous defaults.
type ToolAnnotations struct {
ReadOnlyHint bool `json:"readOnlyHint"`
DestructiveHint bool `json:"destructiveHint"`
}
// Shared annotation presets.
var (
// ReadOnly: pure queries — safe to run without confirmation.
ReadOnly = &ToolAnnotations{ReadOnlyHint: true}
// Mutable: writes state but is not destructive (create/update/register).
Mutable = &ToolAnnotations{}
// Destructive: deletes, reversals, bulk overwrites.
Destructive = &ToolAnnotations{DestructiveHint: true}
)
// AnnotateByPrefix fills in Annotations for tools that don't carry one yet,
// using the repo-wide tool naming convention:
//
// get_*/list_*/search_*/find_*/suggest_*/describe_* → ReadOnly
// delete_*/remove_*/reverse_*/import_*/purge_* → Destructive
// export_*/calculate_*/verify_*/preview_* → ReadOnly (pure computation)
//
// Tools that set Annotations explicitly are left untouched, so a package can
// override individual cases after calling this.
func AnnotateByPrefix(tools []Tool) {
for i := range tools {
if tools[i].Annotations != nil {
continue
}
name := strings.ToLower(tools[i].Name)
switch {
case hasAny(name,
"get_", "list_", "search_", "find_", "suggest_",
"export_", "calculate_", "verify_", "preview_"):
tools[i].Annotations = ReadOnly
case hasAny(name,
"delete_", "remove_", "reverse_", "import_", "purge_"):
tools[i].Annotations = Destructive
default:
tools[i].Annotations = Mutable
}
}
}
func hasAny(name string, prefixes ...string) bool {
for _, p := range prefixes {
if strings.HasPrefix(name, p) {
return true
}
}
return false
}
// Decode unmarshals tool arguments into v, treating empty/null args as a no-op // Decode unmarshals tool arguments into v, treating empty/null args as a no-op
// (zero value). Returns a wrapped error suitable for returning from Invoke. // (zero value). Returns a wrapped error suitable for returning from Invoke.
func Decode(args json.RawMessage, v any) error { func Decode(args json.RawMessage, v any) error {
+146
View File
@@ -0,0 +1,146 @@
// Package mcpauth guards the MCP HTTP endpoints (cart-mcp, order-mcp,
// wishlist-mcp, promotions-mcp, ledger-mcp, finder-mcp) with authentication.
//
// These are admin tooling surfaces for AI agents / operator tooling, so the
// default is FAIL-CLOSED: when no credential source is configured the
// endpoint returns 503 rather than serving unauthenticated.
//
// Credential sources, checked in order (first match wins):
//
// 1. Minted API keys — AUTH_APIKEYS_FILE points at the shared key store
// (the same /commerce/apikeys.json the backoffice and CMS use). Keys
// must be minted with an "mcp" or "admin" scope. The store is reloaded
// on mtime change, so a newly minted key works without a restart.
// Presented as: X-API-Key: <key>
// 2. Device access tokens — AUTH_TOKEN_SECRET enables the platform
// device-authorization flow's local JWTs (HS256). The token's scopes
// must include "mcp" or "admin". Presented as: Authorization: Bearer <jwt>
// 3. Legacy shared key — MCP_API_KEY (dev compose sets devkey; production
// reads it from the stack-secrets secret). Presented as:
// X-API-Key: <key> or Authorization: Bearer <key>
package mcpauth
import (
"crypto/subtle"
"net/http"
"os"
"strings"
"sync"
"git.k6n.net/mats/platform/auth"
)
// authz is the combined validator built from the configured sources.
type authz struct {
keys *auth.KeyStore // minted keys (AUTH_APIKEYS_FILE); nil if unset
jwt *auth.LocalJWT // device tokens (AUTH_TOKEN_SECRET); nil if unset
legacyKey string // MCP_API_KEY; empty if unset
configured bool // at least one source enabled
}
var (
authzOnce sync.Once
current *authz
)
// buildAuthz initializes the validator from the environment once. Env is
// read at first use (Middleware construction), which happens at service
// startup.
func buildAuthz() *authz {
authzOnce.Do(func() {
a := &authz{}
if p := os.Getenv("AUTH_APIKEYS_FILE"); p != "" {
if ks, err := auth.OpenKeyStore(p); err == nil {
a.keys = ks
a.configured = true
}
}
if s := os.Getenv("AUTH_TOKEN_SECRET"); s != "" {
a.jwt = auth.NewLocalJWT(s)
a.configured = true
}
if k := os.Getenv("MCP_API_KEY"); k != "" {
a.legacyKey = k
a.configured = true
}
current = a
})
return current
}
// Middleware protects the wrapped MCP handler. See the package comment for
// the credential contract. The same source order applies; a request fails
// with 401 unless exactly one source validates it.
func Middleware(next http.Handler) http.Handler {
a := buildAuthz()
return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if !a.configured {
http.Error(w, "mcp: no credential source configured (set AUTH_APIKEYS_FILE, AUTH_TOKEN_SECRET, or MCP_API_KEY)", http.StatusServiceUnavailable)
return
}
// Reload the key store when the shared file changed (cheap stat).
if a.keys != nil {
_ = a.keys.ReloadIfChanged()
}
if !a.authenticate(r) {
http.Error(w, "unauthorized", http.StatusUnauthorized)
return
}
next.ServeHTTP(w, r)
})
}
// authenticate tries each configured source and applies scope gating (mcp or
// admin required; a plain read/write-only key is not enough for admin MCP
// surfaces).
func (a *authz) authenticate(r *http.Request) bool {
p, err := a.principal(r)
if err != nil {
return false
}
for _, s := range p.Scopes {
if s == "mcp" || s == "admin" {
return true
}
}
return false
}
// principal resolves the request to a Principal via the configured sources,
// in order: minted key → device token → legacy key. A source that rejects a
// presented credential falls through to the next one, so a shared dev key
// still works when a keystore is also configured.
func (a *authz) principal(r *http.Request) (*auth.Principal, error) {
key := r.Header.Get("X-API-Key")
bearer, hasBearer := strings.CutPrefix(r.Header.Get("Authorization"), "Bearer ")
var lastErr error
if key != "" && a.keys != nil {
if p, err := (auth.APIKeyAuth{Keys: a.keys}).Authenticate(r); err == nil {
return p, nil
} else {
lastErr = err
}
}
if hasBearer && a.jwt != nil {
if p, err := a.jwt.Authenticate(r); err == nil {
return p, nil
} else {
lastErr = err
}
}
// Legacy: accept the shared key via either header form.
if a.legacyKey != "" {
presented := key
if presented == "" && hasBearer {
presented = bearer
}
if presented != "" && subtle.ConstantTimeCompare([]byte(presented), []byte(a.legacyKey)) == 1 {
return &auth.Principal{Subject: "legacy-mcp", Scopes: []string{"mcp"}, Kind: auth.KindAPIKey}, nil
}
}
if lastErr != nil {
return nil, lastErr
}
return nil, auth.ErrNoCredentials
}
+233
View File
@@ -0,0 +1,233 @@
package mcpauth
import (
"net/http"
"net/http/httptest"
"os"
"path/filepath"
"strings"
"sync"
"testing"
"git.k6n.net/mats/platform/auth"
)
// freshAuthz resets the package-level singleton so each test can configure
// its own env. buildAuthz is sync.Once'd, so tests mutate the env BEFORE the
// first Middleware() call.
func resetAuthz(t *testing.T) {
t.Helper()
for _, k := range []string{"AUTH_APIKEYS_FILE", "AUTH_TOKEN_SECRET", "MCP_API_KEY"} {
_ = os.Unsetenv(k)
}
authzOnce = sync.Once{}
current = nil
t.Cleanup(func() {
authzOnce = sync.Once{}
current = nil
})
}
func doReq(t *testing.T, r *http.Request) int {
t.Helper()
rr := httptest.NewRecorder()
Middleware(http.HandlerFunc(func(w http.ResponseWriter, _ *http.Request) {
w.WriteHeader(http.StatusOK)
})).ServeHTTP(rr, r)
return rr.Code
}
func TestNoSourceConfigured(t *testing.T) {
resetAuthz(t)
r := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
if code := doReq(t, r); code != http.StatusServiceUnavailable {
t.Fatalf("no source configured: got %d, want 503", code)
}
}
func TestMintedKeyRequiresMCPOrAdminScope(t *testing.T) {
resetAuthz(t)
path := filepath.Join(t.TempDir(), "apikeys.json")
ks, err := auth.OpenKeyStore(path)
if err != nil {
t.Fatal(err)
}
// Keys with mcp / admin scopes validate.
mcpKey, _, err := ks.Mint("companion", []string{"mcp"})
if err != nil {
t.Fatal(err)
}
adminKey, _, err := ks.Mint("ops", []string{"admin"})
if err != nil {
t.Fatal(err)
}
// A read-only key must NOT validate against MCP surfaces.
readKey, _, err := ks.Mint("reader", []string{"read"})
if err != nil {
t.Fatal(err)
}
_ = os.Setenv("AUTH_APIKEYS_FILE", path)
cases := []struct {
name string
key string
want int
}{
{"mcp scope", mcpKey, http.StatusOK},
{"admin scope", adminKey, http.StatusOK},
{"read-only rejected", readKey, http.StatusUnauthorized},
{"garbage rejected", "gck_bogus", http.StatusUnauthorized},
}
for _, tc := range cases {
t.Run(tc.name, func(t *testing.T) {
r := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r.Header.Set("X-API-Key", tc.key)
if code := doReq(t, r); code != tc.want {
t.Fatalf("got %d, want %d", code, tc.want)
}
})
}
}
func TestKeyStoreReloadPicksUpNewKeys(t *testing.T) {
resetAuthz(t)
path := filepath.Join(t.TempDir(), "apikeys.json")
ks, err := auth.OpenKeyStore(path)
if err != nil {
t.Fatal(err)
}
firstKey, _, err := ks.Mint("a", []string{"mcp"})
if err != nil {
t.Fatal(err)
}
_ = os.Setenv("AUTH_APIKEYS_FILE", path)
r := httptest.NewRequest(http.MethodGet, "/order-mcp", nil)
r.Header.Set("X-API-Key", firstKey)
if code := doReq(t, r); code != http.StatusOK {
t.Fatalf("first key should validate: got %d", code)
}
// A second key minted into the same file AFTER the middleware loaded
// must validate via the mtime-based reload.
secondKey, _, err := ks.Mint("b", []string{"mcp"})
if err != nil {
t.Fatal(err)
}
r2 := httptest.NewRequest(http.MethodGet, "/order-mcp", nil)
r2.Header.Set("X-API-Key", secondKey)
if code := doReq(t, r2); code != http.StatusOK {
t.Fatalf("newly minted key should validate after reload: got %d", code)
}
}
func TestDeviceToken(t *testing.T) {
resetAuthz(t)
_ = os.Setenv("AUTH_TOKEN_SECRET", "test-secret")
jwt := auth.NewLocalJWT("test-secret")
tok, err := jwt.Mint("user-1", []string{"mcp"}, 0) // ttl 0 → exp now (still valid)
if err != nil {
t.Fatal(err)
}
r := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r.Header.Set("Authorization", "Bearer "+tok)
if code := doReq(t, r); code != http.StatusOK {
t.Fatalf("device token with mcp scope: got %d, want 200", code)
}
// A device token without mcp/admin scope must be rejected.
tokNoScope, err := jwt.Mint("user-2", []string{"read"}, 0)
if err != nil {
t.Fatal(err)
}
r2 := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r2.Header.Set("Authorization", "Bearer "+tokNoScope)
if code := doReq(t, r2); code != http.StatusUnauthorized {
t.Fatalf("device token without mcp scope: got %d, want 401", code)
}
// Garbage bearer with AUTH_TOKEN_SECRET set must be 401 (not 503).
r3 := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r3.Header.Set("Authorization", "Bearer not.a.jwt")
if code := doReq(t, r3); code != http.StatusUnauthorized {
t.Fatalf("garbage token: got %d, want 401", code)
}
}
func TestLegacyKey(t *testing.T) {
resetAuthz(t)
_ = os.Setenv("MCP_API_KEY", "devkey")
for _, hdr := range []string{"X-API-Key", "Authorization"} {
r := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
if hdr == "Authorization" {
r.Header.Set(hdr, "Bearer devkey")
} else {
r.Header.Set(hdr, "devkey")
}
if code := doReq(t, r); code != http.StatusOK {
t.Fatalf("legacy key via %s: got %d, want 200", hdr, code)
}
}
r := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r.Header.Set("X-API-Key", "wrong")
if code := doReq(t, r); code != http.StatusUnauthorized {
t.Fatalf("wrong legacy key: got %d, want 401", code)
}
}
func TestSourcesCompose(t *testing.T) {
resetAuthz(t)
path := filepath.Join(t.TempDir(), "apikeys.json")
ks, err := auth.OpenKeyStore(path)
if err != nil {
t.Fatal(err)
}
key, _, err := ks.Mint("companion", []string{"mcp"})
if err != nil {
t.Fatal(err)
}
_ = os.Setenv("AUTH_APIKEYS_FILE", path)
_ = os.Setenv("AUTH_TOKEN_SECRET", "test-secret")
_ = os.Setenv("MCP_API_KEY", "devkey")
// Minted key wins.
r := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r.Header.Set("X-API-Key", key)
if code := doReq(t, r); code != http.StatusOK {
t.Fatalf("minted key: got %d, want 200", code)
}
// Device token works alongside.
jwt := auth.NewLocalJWT("test-secret")
tok, err := jwt.Mint("u", []string{"admin"}, 0)
if err != nil {
t.Fatal(err)
}
r2 := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r2.Header.Set("Authorization", "Bearer "+tok)
if code := doReq(t, r2); code != http.StatusOK {
t.Fatalf("device token: got %d, want 200", code)
}
// Legacy key still works.
r3 := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
r3.Header.Set("X-API-Key", "devkey")
if code := doReq(t, r3); code != http.StatusOK {
t.Fatalf("legacy key: got %d, want 200", code)
}
// Without credentials → 401.
r4 := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
if code := doReq(t, r4); code != http.StatusUnauthorized {
t.Fatalf("no creds: got %d, want 401", code)
}
}
// Ensure the fail-closed behavior matches the documented contract even when
// the env contains empty-but-set values.
func TestEmptyEnvFallsClosed(t *testing.T) {
resetAuthz(t)
_ = os.Setenv("AUTH_APIKEYS_FILE", strings.Repeat(" ", 0))
r := httptest.NewRequest(http.MethodGet, "/cart-mcp", nil)
if code := doReq(t, r); code != http.StatusServiceUnavailable {
t.Fatalf("empty env should fall closed: got %d, want 503", code)
}
}