MCP server for cgx — JSON-RPC stdio server for AI editor integrations. Provides knowledge graph analysis of codebases via Tree-sitter parsing, DuckDB storage, and git analysis.
cgx provides 11 well-named, read-only tools with clear verb-noun naming (get_*, find_*, search_*). All tools have descriptions (avg ~80 chars, within baseline 34-392). Input schemas are present and properly typed with JSON Schema. However, parameter descriptions are sparse or missing entirely, most params lack guidance on format, constraints, or valid ranges. Output schemas are not documented, forcing LLMs to infer result structure. Error handling is minimal; tools return generic error strings without recovery guidance. No tool annotations (readOnlyHint, etc.) despite all being read-only. Composition is sound, tools chain well (e.g., find_symbol → get_neighbors → get_call_chain). Security is appropriate for read-only operations.
Find any function, class, variable, or type by name. Supports optional kind filter.
Find all nodes affected if the given node changes. Returns affected nodes with risk level.
Trace a call path between two symbols. Returns the path if found.
Get all nodes in a community cluster.
Detect unreferenced exports, unused functions, zombie files, and other dead code with confidence scoring.
Get git blame ownership breakdown for a file.
Get highest-risk files by churn × coupling score.
Parameter descriptions missing or incomplete. 'kind', 'depth', 'node_id', 'community_id', 'limit', 'confidence', 'path' lack guidance on valid values, ranges, or formats. LLMs cannot infer constraints from names alone.
Output schemas not documented. Tools return JSON but LLMs have no formal schema for result structure. Clients must reverse-engineer response format from examples.
No tool annotations despite all tools being read-only. Missing readOnlyHint would help clients optimize caching and retry logic.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Get direct dependencies of a node (nodes that it calls or imports).
Get a full architectural overview of the indexed codebase. Always call this first in a new session.
Run a read-only SQL query against the knowledge graph database.
Full-text search over node names and paths.
Error handling returns generic error strings without recovery guidance. 'Unknown tool' or database errors do not tell LLM what to do next or whether to retry.
Parameter 'kind' in find_symbol and get_dead_code lacks enum constraint. Description says 'Optional kind filter: Function, Class, File, Type, Variable' but no formal enum in schema forces LLM to guess valid values.