# graph-tool-call MCP server

> Lets an AI client search large tool catalogs and fetch only the target tools and prerequisite producers it needs to act.

- Repository: https://github.com/SonAIengine/graph-tool-call
- Page: https://hostedmcp.org/servers/sonaiengine/graph-tool-call
- Transport: —
- Stars: 18
- License: MIT

## Tools (3)

- `evidence`
- `token`
- `side`

## FAQ

### How does graph-tool-call reduce the tools sent to a model?

It retrieves the target tool plus typed prerequisite producers and admits only token-budgeted, contract-projected schemas to the model-facing catalog.

### Which tool catalog formats can it ingest?

It ingests OpenAPI, GraphQL introspection, MCP tools, Python functions, and structured tool catalogs.

### Does it run as an MCP server?

Yes, it can serve a built graph over Streamable HTTP at the /mcp endpoint, and a proxy mode puts many MCP backends behind three gateway tools.

### Does it require extra Python dependencies?

The core search and graph package uses only the Python standard library; integrations like OpenAPI, LangChain, and MCP are optional extras.

### Does it handle authentication and authorization?

No, it is the retrieval and contract layer; the host application owns authentication, tenant policy, and approval.