codicil — Documentation Indexing MCP¶
Repo documentation made queryable through MCP. codicil indexes your repository's Markdown, YAML, and TOML documentation into a local Chroma vector store. AI agents can then query docs semantically. Uses Ollama embeddings when available; gracefully degrades to keyword search with zero external dependencies.
What It Does for Operators¶
- Semantic doc search — AI agents can query your project's documentation using natural language
- Zero infra — No external API keys needed. Works entirely locally with Chroma (embedded) + optional Ollama
- Graceful degradation — If Ollama isn't running, falls back to live keyword search on disk — no failure mode
- Auto-reindex — Reindex command for when docs change
- Multi-format — Supports Markdown, YAML, and TOML documentation files
Installation¶
git clone https://github.com/colehellman/codicil.git
cd codicil
pip install -r requirements.txt
# Optional: Install Ollama for semantic search
# curl -fsSL https://ollama.com/install.sh | sh
# ollama pull nomic-embed-text
Claude Desktop / Hermes Config¶
{
"mcpServers": {
"codicil": {
"command": "python",
"args": ["-m", "codicil.server"],
"env": {
"CODICIL_DOCS_PATH": "/path/to/your/repo/docs"
}
}
}
}
Key Tools¶
| Tool | Description |
|---|---|
query_docs |
Semantic (or keyword) search across indexed documentation |
reindex_docs |
Rebuild the vector index from updated docs |
get_doc_context |
Retrieve the full document for a search result |
list_indexed |
List all indexed documents and their status |
Note: Tool names are approximate. See github.com/colehellman/codicil.
Operator Use Cases¶
- Development Teams — "How do we handle error retries in our SDK?" → AI agent queries internal docs
- Technical Writers — Verify documentation coverage by querying for missing topics
- Onboarding — New team members ask AI agents about internal processes instead of digging through wikis
- Open Source Maintainers — Give contributors AI-powered documentation search
CorpusIQ Angle¶
codicil solves an internal documentation problem that many CorpusIQ operators face: "where is that documented?" For operators building on CorpusIQ, codicil can index the CorpusIQ docs alongside their own internal documentation, giving AI agents a unified knowledge surface.
Limitations¶
- Chroma is embedded (not production-scale for massive doc corpuses)
- Semantic search quality depends on Ollama embedding model
- No cloud sync — each developer runs their own index
- Python-only (no Node/TypeScript version)