feat(cli): mcp serve curated stdio MCP server
`weknora mcp serve` — long-lived stdio MCP (Model Context Protocol) transport that exposes a fixed, curated tool surface to MCP-aware agents (Claude Desktop, Claude Code, custom MCP clients). Curated tool set (readonly by default): - whoami — active context + tenant - search (hybrid retrieval against a KB) - kb list / view - doc list / view - agent list / view / invoke - session list / view The list is intentionally narrow to the read + agent-invoke surface; destructive verbs (`delete` / `empty` / `upload`) are gated behind `--write`. Schema is built from each leaf cobra command's flags so adding a new tool is a single registry entry plus a Service interface. Includes the simplify post-review polish + a second simplify pass to fold the resulting feedback (typed schemas, agent_help wording, unify chat / agent invoke option names).
This commit is contained in:
+29
-112
@@ -13,14 +13,15 @@ import (
|
||||
"github.com/Tencent/WeKnora/cli/internal/cmdutil"
|
||||
"github.com/Tencent/WeKnora/cli/internal/format"
|
||||
"github.com/Tencent/WeKnora/cli/internal/iostreams"
|
||||
"github.com/Tencent/WeKnora/cli/internal/sse"
|
||||
sdk "github.com/Tencent/WeKnora/client"
|
||||
)
|
||||
|
||||
// agentInvokeFields enumerates fields surfaced for `--json` discovery on
|
||||
// `agent invoke`. Matches invokeData below — single-shot result envelope
|
||||
// with the agent's final answer plus the trace (references, tool calls).
|
||||
// with the agent's final answer plus the trace (references, tool events).
|
||||
var agentInvokeFields = []string{
|
||||
"answer", "references", "tool_calls", "thinking",
|
||||
"answer", "references", "tool_events", "thinking",
|
||||
"session_id", "agent_id", "query",
|
||||
}
|
||||
|
||||
@@ -45,25 +46,15 @@ type InvokeService interface {
|
||||
AgentQAStreamWithRequest(ctx context.Context, sessionID string, req *sdk.AgentQARequest, cb sdk.AgentEventCallback) error
|
||||
}
|
||||
|
||||
// toolCallTrace mirrors a single SSE `tool_call` event so agents can see
|
||||
// which tools the WeKnora-side agent invoked (and their results). Only the
|
||||
// fields the server actually emits are captured.
|
||||
type toolCallTrace struct {
|
||||
ID string `json:"id"`
|
||||
Name string `json:"name,omitempty"`
|
||||
Result string `json:"result,omitempty"`
|
||||
Data map[string]any `json:"data,omitempty"`
|
||||
}
|
||||
|
||||
// invokeData is the JSON envelope payload.
|
||||
type invokeData struct {
|
||||
Answer string `json:"answer"`
|
||||
References []*sdk.SearchResult `json:"references"`
|
||||
ToolCalls []toolCallTrace `json:"tool_calls,omitempty"`
|
||||
Thinking string `json:"thinking,omitempty"`
|
||||
SessionID string `json:"session_id"`
|
||||
AgentID string `json:"agent_id"`
|
||||
Query string `json:"query"`
|
||||
Answer string `json:"answer"`
|
||||
References []*sdk.SearchResult `json:"references"`
|
||||
ToolEvents []sse.AgentToolEvent `json:"tool_events,omitempty"`
|
||||
Thinking string `json:"thinking,omitempty"`
|
||||
SessionID string `json:"session_id"`
|
||||
AgentID string `json:"agent_id"`
|
||||
Query string `json:"query"`
|
||||
}
|
||||
|
||||
// NewCmdInvoke builds `weknora agent invoke <agent-id> "<text>"`.
|
||||
@@ -152,12 +143,12 @@ func runInvoke(ctx context.Context, opts *InvokeOptions, jopts *cmdutil.JSONOpti
|
||||
Channel: "api",
|
||||
}
|
||||
|
||||
acc := newAgentAccumulator()
|
||||
acc := &sse.AgentAccumulator{}
|
||||
cb := func(r *sdk.AgentStreamResponse) error {
|
||||
if streamMode && r != nil && r.ResponseType == sdk.AgentResponseTypeAnswer && r.Content != "" {
|
||||
_, _ = iostreams.IO.Out.Write([]byte(r.Content))
|
||||
}
|
||||
acc.append(r)
|
||||
acc.Append(r)
|
||||
return nil
|
||||
}
|
||||
|
||||
@@ -169,7 +160,7 @@ func runInvoke(ctx context.Context, opts *InvokeOptions, jopts *cmdutil.JSONOpti
|
||||
if errors.Is(streamErr, context.Canceled) || errors.Is(ctx.Err(), context.Canceled) {
|
||||
return cmdutil.Wrapf(cmdutil.CodeUserAborted, streamErr, "agent invoke cancelled")
|
||||
}
|
||||
if acc.answer.Len() > 0 && !acc.done {
|
||||
if acc.Answer() != "" && !acc.Done() {
|
||||
return cmdutil.Wrapf(cmdutil.CodeSSEStreamAborted, streamErr, "stream aborted before completion")
|
||||
}
|
||||
return cmdutil.WrapHTTP(streamErr, "agent-chat stream")
|
||||
@@ -177,17 +168,17 @@ func runInvoke(ctx context.Context, opts *InvokeOptions, jopts *cmdutil.JSONOpti
|
||||
|
||||
// Server closed cleanly but never sent a Done event — treat as aborted
|
||||
// so agents don't silently emit a truncated answer as ok=true.
|
||||
if !acc.done {
|
||||
if !acc.Done() {
|
||||
return cmdutil.NewError(cmdutil.CodeSSEStreamAborted, "stream ended without a terminal event")
|
||||
}
|
||||
|
||||
answer := acc.answer.String()
|
||||
answer := acc.Answer()
|
||||
if jsonOut {
|
||||
data := invokeData{
|
||||
Answer: answer,
|
||||
References: acc.references,
|
||||
ToolCalls: acc.toolCalls,
|
||||
Thinking: acc.thinking.String(),
|
||||
References: acc.References,
|
||||
ToolEvents: acc.ToolEvents,
|
||||
Thinking: acc.Thinking(),
|
||||
SessionID: sessionID,
|
||||
AgentID: opts.AgentID,
|
||||
Query: opts.Query,
|
||||
@@ -210,98 +201,24 @@ func runInvoke(ctx context.Context, opts *InvokeOptions, jopts *cmdutil.JSONOpti
|
||||
fmt.Fprintln(out)
|
||||
}
|
||||
}
|
||||
renderToolTrace(out, acc.toolCalls)
|
||||
renderReferences(out, acc.references)
|
||||
renderToolTrace(out, acc.ToolEvents)
|
||||
format.WriteReferences(out, acc.References)
|
||||
return nil
|
||||
}
|
||||
|
||||
// agentAccumulator buffers an AgentQAStream callback sequence. Distinct
|
||||
// from internal/sse.Accumulator because the agent event model is wider
|
||||
// (thinking / tool_call / tool_result / answer / references / reflection /
|
||||
// error) and uses a flat r.Done bool instead of the
|
||||
// ResponseType=complete sentinel that KnowledgeQAStream emits.
|
||||
type agentAccumulator struct {
|
||||
answer strings.Builder
|
||||
thinking strings.Builder
|
||||
references []*sdk.SearchResult
|
||||
toolCalls []toolCallTrace
|
||||
done bool
|
||||
}
|
||||
|
||||
func newAgentAccumulator() *agentAccumulator { return &agentAccumulator{} }
|
||||
|
||||
func (a *agentAccumulator) append(r *sdk.AgentStreamResponse) {
|
||||
if r == nil || a.done {
|
||||
return
|
||||
}
|
||||
switch r.ResponseType {
|
||||
case sdk.AgentResponseTypeAnswer:
|
||||
if r.Content != "" {
|
||||
a.answer.WriteString(r.Content)
|
||||
}
|
||||
case sdk.AgentResponseTypeThinking, sdk.AgentResponseTypeReflection:
|
||||
if r.Content != "" {
|
||||
a.thinking.WriteString(r.Content)
|
||||
}
|
||||
case sdk.AgentResponseTypeReferences:
|
||||
if r.KnowledgeReferences != nil {
|
||||
a.references = r.KnowledgeReferences
|
||||
}
|
||||
case sdk.AgentResponseTypeToolCall, sdk.AgentResponseTypeToolResult:
|
||||
a.toolCalls = append(a.toolCalls, toolCallTrace{
|
||||
ID: r.ID,
|
||||
Name: string(r.ResponseType),
|
||||
Result: r.Content,
|
||||
Data: r.Data,
|
||||
})
|
||||
}
|
||||
// References can also arrive on the terminal frame.
|
||||
if r.KnowledgeReferences != nil && a.references == nil {
|
||||
a.references = r.KnowledgeReferences
|
||||
}
|
||||
if r.Done {
|
||||
a.done = true
|
||||
}
|
||||
}
|
||||
|
||||
// renderToolTrace prints a compact tool-call footer in human mode. Skipped
|
||||
// when the agent invoked no tools — silent beats an empty banner.
|
||||
func renderToolTrace(w io.Writer, calls []toolCallTrace) {
|
||||
if len(calls) == 0 {
|
||||
// renderToolTrace prints a compact tool-event footer in human mode.
|
||||
// Skipped when the agent emitted no tool events — silent beats an empty
|
||||
// banner.
|
||||
func renderToolTrace(w io.Writer, events []sse.AgentToolEvent) {
|
||||
if len(events) == 0 {
|
||||
return
|
||||
}
|
||||
fmt.Fprintln(w)
|
||||
fmt.Fprintln(w, "──── Tool trace ────")
|
||||
for i, c := range calls {
|
||||
fmt.Fprintf(w, "[%d] %s", i+1, c.Name)
|
||||
if c.Result != "" {
|
||||
fmt.Fprintf(w, " %s", truncateInline(c.Result, 80))
|
||||
}
|
||||
fmt.Fprintln(w)
|
||||
}
|
||||
}
|
||||
|
||||
// renderReferences mirrors chat.go's references footer for parity.
|
||||
func renderReferences(w io.Writer, refs []*sdk.SearchResult) {
|
||||
if len(refs) == 0 {
|
||||
return
|
||||
}
|
||||
fmt.Fprintln(w)
|
||||
fmt.Fprintln(w, "──── References ────")
|
||||
for i, r := range refs {
|
||||
if r == nil {
|
||||
continue
|
||||
}
|
||||
title := r.KnowledgeTitle
|
||||
if title == "" {
|
||||
title = r.KnowledgeFilename
|
||||
}
|
||||
if title == "" {
|
||||
title = r.KnowledgeID
|
||||
}
|
||||
fmt.Fprintf(w, "[%d] %s", i+1, title)
|
||||
if r.Score > 0 {
|
||||
fmt.Fprintf(w, " score=%.3f", r.Score)
|
||||
for i, e := range events {
|
||||
fmt.Fprintf(w, "[%d] %s", i+1, e.Kind)
|
||||
if e.Result != "" {
|
||||
fmt.Fprintf(w, " %s", truncateInline(e.Result, 80))
|
||||
}
|
||||
fmt.Fprintln(w)
|
||||
}
|
||||
|
||||
@@ -150,7 +150,7 @@ func (c *createSessionTracker) CreateSession(ctx context.Context, req *sdk.Creat
|
||||
return c.InvokeService.CreateSession(ctx, req)
|
||||
}
|
||||
|
||||
func TestInvoke_ToolCallsCaptured(t *testing.T) {
|
||||
func TestInvoke_ToolEventsCaptured(t *testing.T) {
|
||||
out, _ := iostreams.SetForTest(t)
|
||||
svc := &scriptedInvokeSvc{events: []*sdk.AgentStreamResponse{
|
||||
toolCallEvent("call_1", "knowledge_search"),
|
||||
@@ -167,11 +167,11 @@ func TestInvoke_ToolCallsCaptured(t *testing.T) {
|
||||
if err := json.Unmarshal(out.Bytes(), &env); err != nil {
|
||||
t.Fatalf("parse: %v", err)
|
||||
}
|
||||
if len(env.Data.ToolCalls) != 1 {
|
||||
t.Fatalf("expected 1 tool call, got %d", len(env.Data.ToolCalls))
|
||||
if len(env.Data.ToolEvents) != 1 {
|
||||
t.Fatalf("expected 1 tool call, got %d", len(env.Data.ToolEvents))
|
||||
}
|
||||
if env.Data.ToolCalls[0].ID != "call_1" {
|
||||
t.Errorf("tool_calls[0].id = %q, want call_1", env.Data.ToolCalls[0].ID)
|
||||
if env.Data.ToolEvents[0].ID != "call_1" {
|
||||
t.Errorf("tool_calls[0].id = %q, want call_1", env.Data.ToolEvents[0].ID)
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user