SonAIengine/graph-tool-call

Overview

A Python library for graph-based tool retrieval in LLM agents. It selects relevant tools from a set of up to 248 tools using a graph structure, achieving 82% accuracy while consuming 79% fewer tokens than baseline approaches. The library has zero external dependencies and supports OpenAPI, MCP, and LangChain integrations.

Best for

Best for
Developers building LLM agents that need efficient tool selection from a large tool set

Use cases

• Selecting tools for LLM agents with minimal token overhead
• Integrating graph-based tool retrieval into agent workflows via OpenAPI or MCP
• Reducing context lengths when managing large tool inventories

How to use

Install

uvx graph-tool-call search "user authentication" \

Tools exposed

• embedding-local

Tested with

Claude Code, Cursor, Windsurf, ChatGPT

Notes

A Python library for graph-based tool retrieval in LLM agents. It selects relevant tools from a set of up to 248 tools using a graph structure, achieving 82% accuracy while consuming 79% fewer tokens than baseline approaches. The library has zero external dependencies and supports OpenAPI, MCP, and LangChain integrations.

7 stars on GitHub. Last updated 2026-05-06. Licensed MIT.

Use cases

• Selecting tools for LLM agents with minimal token overhead
• Integrating graph-based tool retrieval into agent workflows via OpenAPI or MCP
• Reducing context lengths when managing large tool inventories

Pros

• Zero dependencies, lightweight and easy to embed
• Substantial token savings (79%) without major accuracy tradeoff
• Supports multiple standard interfaces (OpenAPI, MCP, LangChain)

Cons

• Limited community adoption (7 GitHub stars)
• Python-only, not available for other languages
• 82% accuracy on 248 tools may degrade with much larger tool sets

Indexed from awesome-mcp-servers-punkpeye and enriched against its public facts.