Token-efficient code intelligence MCP server — indexes a codebase with tree-sitter and lets AI agents fetch exactly the symbols they need
pitlane-mcp exposes 25 tools with inconsistent schema documentation. Tool names follow verb_noun convention (search_symbols, get_file_outline, find_callers) which is good. However, critical issues: (1) Input schemas are inferred from parameter descriptions in Python benchmark code, not visible in actual Rust source (src/main.rs shows only tool registration without explicit schema definitions). (2) Most tool descriptions are 1-2 sentences (50-100 chars), below the 194-char baseline for A+ tools. (3) Parameter descriptions exist but lack type constraints, ranges, and format specifications. (4) No documented output schemas, LLMs cannot predict response structure. (5) Error handling and recovery guidance absent. (6) No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tools like ensure_project_ready and watch_project are WRITE operations but lack explicit state-change warnings in descriptions.
Analyze the impact of changes to a symbol or file on the rest of the codebase.
Ensure a project is indexed and ready for querying. Optionally starts a background watcher for incremental updates.
Find all functions or methods called by a specific symbol.
Find all functions or methods that call a specific symbol.
Find all usages of a specific symbol across the codebase.
Get the symbol outline (tree of definitions) for a specific file.
Get a feed of symbols changed since an index revision.
Input schemas not visible in source code. Tool definitions inferred from Python benchmark harness, not from Rust implementation. Cannot verify actual parameter types, constraints, or validation.
No documented output schemas. LLMs cannot predict response structure, field names, or types. Prevents downstream tool chaining and forces agents to guess at response format.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Get statistics about the indexed project including symbol counts, languages, and embedding progress.
Retrieve specific lines of source code from a file.
Get a high-level outline of the project structure and top-level symbols.
Retrieve detailed information about a specific symbol by ID.
Get usage statistics for the MCP server.
Index a project and persist the symbol index to disk. Returns immediately after indexing completes.
Investigate a code area by gathering related symbols and context.
Locate code by searching for symbols, files, or patterns.
Navigate through code by following symbol references and definitions.
Read a specific code unit (function, class, module, etc.) with full context.
Read the full content of a knowledge document.
Search for content within source files using BM25 full-text search or exact matching.
Search for files by name or path pattern within the indexed project.
Search knowledge documents (markdown, etc.) with natural language queries.
Search for symbols by name or pattern across the indexed project.
Trace an execution path through the codebase from one symbol to another.
Wait for embeddings to be generated for the indexed project.
Start or manage a background file watcher for incremental index updates.
Tool descriptions are 50-100 characters (below 194-char baseline). Lack WHEN to use, prerequisites, and state-change warnings. E.g., ensure_project_ready and watch_project are WRITE operations but descriptions do not warn of side effects.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Agents cannot distinguish safe read-only tools from destructive ones. ensure_project_ready, index_project, and watch_project modify state but lack destructiveHint.
No error handling or recovery guidance. Tools lack descriptions of failure modes, retryability, or next steps. E.g., search_symbols does not document what happens if project is not indexed, or how to recover.