feat: MCP Phase 1 — Code Intelligence tools (#284)

[ajianaz]

MCP Phase 1 — Code Intelligence Tools

Expose v0.6/v0.7 code intelligence to AI agents (Claude Code, Cursor, Windsurf).

5 New MCP Tools

Tool	Input	Output	Verified
cora.search_symbols	query, kind?, file?, language?, limit?	JSON: matching symbols	✅
cora.find_callers	symbol, limit?	JSON: caller locations	✅ 11 callers
cora.find_impact	symbol, depth?	JSON: affected symbols by depth	✅ 21 sites
cora.find_affected_tests	files[]	JSON: affected test files	✅ 2 files
cora.index_status	—	JSON: symbol/file/language stats	✅ 534 symbols

Design

• Read-only: no side effects, no writes
• JSON output: structured for agent parsing
• Graceful: "no index, run cora index first" error
• Fast: <100ms (no LLM calls)
• Stateless: each call independent

Tests

• 559 total (+7 new: parameter validation + tool listing)
• Clippy clean, fmt clean
• Runtime verified via full MCP protocol handshake + tool calls

Example Agent Session

Agent: "Where is open_index defined?"
→ cora.search_symbols("open_index")
→ [{name: "open_index", kind: "function", file: "src/index/mod.rs", line: 22}]

Agent: "What breaks if I change it?"
→ cora.find_impact("open_index", depth=2)
→ [21 affected symbols across 2 depth levels]

Agent: "Which tests should I run?"
→ cora.find_affected_tests(["src/index/mod.rs"])
→ {affected_tests: ["tests/cli_basic.rs"], count: 1}