helix-docs-agent

Setup

uv venv
source .venv/bin/activate
uv sync
uv run ...

Files

• scrape_data: Scripts to scrape the documentation from the HelixDB docs repository and ingest it into the database.
• /local_models: USES OLLAMA - has a tool calling agent and a summarization agent. Tool calling agent queries HelixDB mcp endpoints and traverses the graph to find relevant files. Summarization agent summarizes the results into a natural language response.
• /openrouter: USES OPENROUTER - has just one agent that can do everything.
• run_mcp.py: MCP server for HelixDB graph traversal and search over STDIO.
• run_mcp_http.py: MCP server for HelixDB graph traversal and search over HTTP.
• search_local.py: Example of how to run a vector search query
• db/queries.hx: HelixDB queries for the documentation database.
• db/schema.hx: HelixDB schema for the documentation database.
• data - has HelixDB docs data

Running MCP servers

HTTP

claude mcp add --transport http helix http://0.0.0.0:8000/mcp

STDIO

• Claude code:

claude mcp add --transport stdio helix-docs python UR_DIRECTORY_OF_THE_MCP_SERVER
# or to use local embeddings with Ollama
claude mcp add --transport stdio helix-docs --env LOCAL=true \
    -- python /Users/putt/Documents/Github/helix-docs-agent/run_mcp.py

Removing the MCP server

claude mcp remove helix

Running demo

uv run scrape_data/ingestion.py
uv run openrouter/openrouter_agent.py