v1.0.1, serve --mcp --no-watch -p <monorepo>, after container restart, the first tools/call is unresponsive; CODEGRAPH_NO_DAEMON=1 can reproduce the recovery process.

[HLanguage]

v1.0.1, serve --mcp --no-watch -p , after container restart, the first tools/call is unresponsive; CODEGRAPH_NO_DAEMON=1 can reproduce the recovery process.

[HLanguage]

Title

serve --mcp --no-watch hangs on first tools/call after container restart; CODEGRAPH_NO_DAEMON=1 restores behavior

---

Environment

Item | Value -- | -- Package | @colbymchenry/codegraph@1.0.1 Node | 22.x (Docker node:22-slim) OS | Linux (Docker container, non-root user) Repo | Large Java monorepo (~745 microservices, 93k+ .java files) Index | Pre-built at $REPO/.codegraph/ (codegraph.db ~5.7GB) Command | codegraph serve --mcp --no-watch -p Transport | stdio, newline-delimited JSON (NDJSON) Related env | CODEGRAPH_NO_WATCHDOG=1 (large-repo watchdog; separate issue)

No JSON-RPC error frame; stdout simply stops producing lines for the pending tools/call id.

---

Minimal reproduction (inside container or host with global CLI)

Prerequisites

npm install -g @colbymchenry/codegraph@1.0.1

export REPO_PATH=/path/to/monorepo # must have .codegraph/codegraph.db

codegraph --version # 1.0.1

A. Reproduce hang (default shared daemon)

# Optional: simulate post-restart stale daemon artifacts (often present after kill -9 / container stop)

rm -f "$REPO_PATH/.codegraph/daemon.sock" "$REPO_PATH/.codegraph/daemon.pid"

codegraph unlock "$REPO_PATH" 2>/dev/null || true

# Spawn MCP server (stdio)

codegraph serve --mcp --no-watch -p "$REPO_PATH"

Send NDJSON on stdin (one JSON object per line):

{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2024-11-05","capabilities":{},"clientInfo":{"name":"repro","version":"1.0.0"}}}

{"jsonrpc":"2.0","method":"notifications/initialized"}

{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}

{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"codegraph_search","arguments":{"query":"OrderController","limit":3}}}

Observe: id 1 and 2 get responses; id 3 never arrives (we tested with 600s–6000s client timeout).

B. Workaround (single-process / no shared daemon)

CODEGRAPH_NO_DAEMON=1 codegraph serve --mcp --no-watch -p "$REPO_PATH"

Same JSON-RPC sequence → id 3 returns normally.

C. Container restart scenario (our production path)

docker restart <container>

# wait until app is up; do NOT run any warmup query

docker exec -e REPO_PATH=/data/repo <container> node /app/mcp-server/test-proxy.mjs

test-proxy.mjs flow: spawn wrapper → initialize → tools/list → tools/call(codegraph_search).

• Fails (hang) when child env omits CODEGRAPH_NO_DAEMON
• Passes when CODEGRAPH_NO_DAEMON=1 is set on the codegraph serve child

---

Integration context (optional)

We wrap codegraph serve in a small Node proxy:

spawn("codegraph", ["serve", "--mcp", "--no-watch", "-p", REPO_PATH], {

stdio: ["pipe", "pipe", "inherit"],

env: { ...process.env, CODEGRAPH_NO_DAEMON: process.env.CODEGRAPH_NO_DAEMON ?? "1" }

});

Without CODEGRAPH_NO_DAEMON=1, external MCP clients (Cursor / Claude Code via .mcp.json) hit the hang on first tool use after container restart.

With CODEGRAPH_NO_DAEMON=1, restart + immediate tools/call is stable.

---

Additional notes

1. Not a slow index load: With CODEGRAPH_NO_DAEMON=1, the same query after restart completes; without it, we see zero stdout for the pending tools/call id over 10+ minutes.
2. Stale daemon files: Our entrypoint cleans dead daemon.pid / daemon.sock when PID is gone; hang still occurs on first tools/call without CODEGRAPH_NO_DAEMON.
3. CLI path: Direct codegraph search / CLI proxy may behave differently; this report is specifically serve --mcp + first tools/call.
4. Related: We also set CODEGRAPH_NO_WATCHDOG=1 for large repos (watchdog killing daemon during long cold queries). That is separate from this hang.

---

Suggested fix direction

• Fix shared-daemon attach/handshake on MCP tools/call after process/container restart, or
• Document that MCP stdio integrations should set CODEGRAPH_NO_DAEMON=1, or
• Auto-fallback to in-process mode when daemon socket exists but is unresponsive

---

Workaround (current)

export CODEGRAPH_NO_DAEMON=1

codegraph serve --mcp --no-watch -p <monorepo>