tooling: wire codegraph MCP, pin the local index out of git

[DealAppSeo]

Description

Wires codegraph as a project-scoped MCP server and pins its local index out of git. Tooling only — no protocol code, no spec, no contracts touched.

Part of a five-repo change closing a decision from repid-engine's reports/2026-08-09/TOOLING_EVAL_MEMORY_GRAPH_ROUTING.md, which verdicted codegraph "Install TODAY" and was never executed.

Fixes # (no issue)

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update (tooling/config only — closest fit; nothing in the protocol surface changes)

How Has This Been Tested?

• codegraph init -y . → 51 files indexed, 682 nodes / 1,870 edges in 420ms.
• MCP invocation codegraph serve --mcp taken from codegraph install --print-config claude, not guessed — there is no bare codegraph mcp subcommand.
• Telemetry disabled in .mcp.json (CODEGRAPH_TELEMETRY=0) and globally via codegraph telemetry off.
• .gitignore pin verified: git status shows no .codegraph entries afterwards.

⚠️ One finding recorded here, deliberately not fixed

A new check:drift suite (landing in the trinity-ecosystem PR) scans live surfaces for [VERIFIED <date>] / MEASURED <date> tags that have outlived their window. It flags one claim in this repo:

README.md:73 — "70 lifetime on-chain reputation writes from the agent economy — real production activity, not synthetic", dated 2026-07-08, now 52 days old against a 45-day window for negative/absolute assertions.

This is a public claim about on-chain activity on a public README. It may well still be true — the check reports staleness, not falsity. But it should be re-probed against the chain before it is repeated anywhere, and then re-dated.

It is not probed in this PR because this change set does not touch chain access, and re-dating a claim without running the probe is precisely the failure the check exists to catch. That work belongs to the new claim-prober agent, or to whoever next has an RPC path open.

Checklist:

• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas (the .gitignore pin carries its reason)
• I have made corresponding changes to the documentation (the index row lives in trinity-ecosystem's PRIOR-WORK-INDEX.md)
• My changes generate no new warnings
• Any dependent changes have been merged and published in downstream modules (none — this repo's change is self-contained; the subagent set is not synced here)

---

Generated by Claude Code