Skip to content

agent_skill_codegraphcontext.md

Latest commit

 

History

History
44 lines (31 loc) · 2.77 KB

agent_skill_codegraphcontext.md

File metadata and controls

44 lines (31 loc) · 2.77 KB

Agent skill: CodeGraphContext (CGC)

This is the published copy of the Cursor agent skill for CodeGraphContext. The same content lives under .cursor/skills/codegraphcontext/SKILL.md for Cursor users who clone the repo; both stay in sync when we update workflows.

# YAML frontmatter (Cursor SKILL.md format)
name: codegraphcontext
description: >-
  Use CodeGraphContext (CGC) to index a repo into a graph DB and query it via CLI
  or MCP. Apply when the user wants semantic/code-graph search, call graphs,
  indexing, or configuring the codegraphcontext MCP server.

When to use this skill

  • Indexing or re-indexing a codebase for AI or CLI queries.
  • Explaining how to install and run cgc, choose a database backend, or wire MCP into an editor.
  • Interpreting CGC tools: codegraph_find_code, analyze_code_relationships, add_code_to_graph, etc.

Core workflow

  1. Install: pip install codegraphcontext (or pipx install codegraphcontext). With uv, uv tool install codegraphcontext or uvx codegraphcontext … is supported. tree-sitter-c-sharp is declared in this project’s pyproject.toml, but some uvx isolated environments have still failed to install it; if you see ModuleNotFoundError: tree_sitter_c_sharp, use e.g. uvx --with tree-sitter-c-sharp codegraphcontext ….
  2. Configure: Optional ~/.codegraphcontext/.env for DEFAULT_DATABASE, Neo4j URI, or Kùzu path. Run cgc doctor if connections fail.
  3. Index: From the repo root, cgc index . (or cgc index --force . to rebuild). Ensure CLI and MCP use the same config so they see the same graph.
  4. Query: CLI (cgc find, cgc query, …) or MCP tools after cgc mcp setup / cgc mcp start in the client config.

MCP setup (short)

  • Run cgc mcp setup and pick your editor (including OpenCode for printed stdio instructions), or add a server entry that runs cgc with args mcp start and the same env as the CLI.
  • OpenCode: OpenCode MCP servers.

Agent behavior

  • Prefer indexing the workspace before deep graph queries if the user has not indexed yet.
  • For fuzzy symbol search on Kùzu/Falkor backends, matching is typo-tolerant (edit distance); on Neo4j, full-text fuzzy uses Lucene-style terms—preserve original casing in queries when fuzziness matters for camelCase symbols.
  • If Repository.path is missing in the DB, those rows are skipped for path matching and a warning is logged when repositories are listed; clean up stale nodes if needed.

References in this repo

  • CLI entry: codegraphcontext.cli.main
  • MCP server and tool wiring: codegraphcontext.server, tool_definitions.py
  • User-facing setup: Setup workflows, MCP guide