| name | cocoindex-code |
| description | Semantic vector search over the codebase — find code by meaning, not exact names. Use when you need to locate implementations without knowing exact function/class names. Triggers on "find code", "search codebase", "locate implementation", "what does X do". |
$cocoindex-code — Semantic Search
Finds code by meaning using local vector embeddings. No API key, works offline.
When to Use This vs Other Tools
| Use cocoindex-code | Use serena instead |
|---|
| "find code that handles rate limiting" | "who calls rateLimit()?" |
| "locate authentication middleware" | "what does AuthService import?" |
| "find all error handling patterns" | "trace path from HTTP handler to DB" |
| Exploring unfamiliar codebase | Tracing known functions |
| Fuzzy conceptual search | Exact symbol traversal |
The One Tool: search
{
"query": "string — natural language or code snippet",
"limit": 5,
"offset": 0,
"refresh_index": true,
"languages": ["python", "typescript"],
"paths": ["src/auth/*"]
}
After getting results: Use the file tool on file_path at start_line–end_line for full context.
Performance Tips
- Set
refresh_index: false for consecutive searches in the same session (no code changes)
- Use
languages filter to scope search (faster for single-language queries)
- Default limit is 5. Increase to 10–20 for broad exploration.
Lifecycle
Index not built yet:
uvx cocoindex-code-mcp-server@latest --data-dir .codex/cocoindex --index
After major changes: Index auto-refreshes on each MCP search call (refresh_index=true default).
Config in .codex/config.toml:
[mcp_servers.cocoindex-code]
command = "uvx"
args = ["cocoindex-code-mcp-server@latest", "--data-dir", ".codex/cocoindex"]
Supported Languages (29)
Python, JavaScript, TypeScript/TSX/JSX, Rust, Go, Java, C, C++, C#, SQL,
Shell/Bash, Markdown, PHP, Lua, Ruby, Swift, Kotlin, Scala, R, HTML, CSS/SCSS,
JSON, YAML/TOML, XML, Solidity, Pascal, Fortran, plain text.
