nullkey
5b07c9ab87
New subtree (chunk list / view / delete) exposes RAG retrieval debugging primitives with SDK-grounded field set (23 Chunk fields). Pagination follows v0.4 canon: --limit / --page-size (1..1000) / --all-pages. - chunk list --doc <id>: enumerate by ChunkIndex (separate from search chunks which is hybrid retrieval; Long help documents the distinction) - chunk view <id>: scope-less render via /chunks/by-id route; full content verbatim - chunk delete <id> --doc <id>: scope-flag + scope-id; L-13 destructive; 404 NOT idempotent; resource.not_found / auth.forbidden / input.confirmation_required typed exit codes documented in Long help MCP server gains chunk_list as 10th curated tool. Schema deliberately exposes only doc_id + limit (no pagination workflow on MCP); response includes truncated_at_limit flag when total > limit. cli/AGENTS.md MCP curation rationale rewritten: curated read-only is a deliberate product call because the server side does not yet enforce per-token scope. When server scope ships, mutation tools can land in the MCP surface. Shared helper cli/internal/text/timeago_string.go (FuzzyAgoStr) extracted from session list during the C2 quality-review pass.
60 lines
2.0 KiB
Go
60 lines
2.0 KiB
Go
// Package mcp wires the curated weknora tool set to an
|
|
// modelcontextprotocol/go-sdk server. RunStdio is the entry point invoked
|
|
// by `weknora mcp serve`.
|
|
//
|
|
// Design notes:
|
|
//
|
|
// - Tool surface is hand-curated rather than auto-derived from the cobra
|
|
// tree (which would expose auth/link/completion/destructive verbs that
|
|
// don't belong on an agent-callable surface).
|
|
// - Long-running tools (chat / agent_invoke) accumulate the LLM SSE
|
|
// stream server-side and return a single CallToolResult - MCP spec
|
|
// 2025-06-18 does not define streamed tool-result content, so the
|
|
// accumulate-and-return pattern is the canonical path.
|
|
// - Handlers receive ctx for cancellation; mid-LLM-stream cancellation
|
|
// propagates to the SDK via context, which closes the SSE connection.
|
|
package mcp
|
|
|
|
import (
|
|
"context"
|
|
"fmt"
|
|
|
|
mcpsdk "github.com/modelcontextprotocol/go-sdk/mcp"
|
|
|
|
"github.com/Tencent/WeKnora/cli/internal/build"
|
|
)
|
|
|
|
// ServiceClient bundles the SDK methods the tool registry needs. *sdk.Client
|
|
// satisfies it; tests substitute a fake to exercise the tool handlers
|
|
// in-process without standing up a real WeKnora server.
|
|
//
|
|
// Embedding the full SDK Client would couple every tool test to every SDK
|
|
// method; declaring the narrow surface here keeps the seam tight.
|
|
type ServiceClient interface {
|
|
knowledgeBaseService
|
|
knowledgeService
|
|
chatService
|
|
agentService
|
|
chunkListService
|
|
}
|
|
|
|
// RunStdio constructs the MCP server, registers the curated 10 tools, and
|
|
// blocks reading JSON-RPC from stdin until the client disconnects or ctx
|
|
// is cancelled. Returns the underlying transport error (if any); the cobra
|
|
// RunE caller maps it through the usual cmdutil exit-code path.
|
|
func RunStdio(ctx context.Context, svc ServiceClient) error {
|
|
v, _, _ := build.Info()
|
|
server := mcpsdk.NewServer(
|
|
&mcpsdk.Implementation{
|
|
Name: "weknora",
|
|
Version: v,
|
|
},
|
|
nil,
|
|
)
|
|
registerTools(server, svc)
|
|
if err := server.Run(ctx, &mcpsdk.StdioTransport{}); err != nil {
|
|
return fmt.Errorf("mcp serve: %w", err)
|
|
}
|
|
return nil
|
|
}
|