MCP Servers

MCP Servers

The MCP Servers feature lets you connect your Toolhouse workers to external Model Context Protocol (MCP) servers. This extends your worker's capabilities beyond Toolhouse's built-in integrations by giving it access to tools hosted on your own MCP servers.

Toolhouse is compatible with SSE and Streamable HTTP servers.

You can use MCP servers alongside existing integrations. Your worker can mix and match tools from your MCP servers and from Toolhouse's integrations to complete its tasks.

How your worker uses MCP servers

When your worker runs, it will check whether there are any tools available in your connected MCP servers, and use them alongside any other integrations it has been configured with. From the worker's perspective, MCP server tools and Toolhouse integrations work together as part of the same toolset.

Adding MCP servers can significantly increase the context your worker has to process. If a server exposes many tools, this can affect your worker's performance.

MCP can be context heavy. If you encounter issues, you can edit your worker to remove some of your integrations. Alternatively, you can create a separate worker that specifically handles the MCP server, and have an orchestrator worker call your MCP-enabled worker. See Worker Orchestration for more details.

• Guide your worker explicitly. Because Toolhouse does not know the details of your MCP server, you should explicitly guide your worker to use the right tools provided by the server. This ensures the best performance.

• No code interpreter support. Currently, MCP servers cannot be called from within the code interpreter.

Adding an MCP server

• Go to Agents in your Toolhouse

• Click on your worker to edit it

• Select MCP Servers

• Add the full URL to the MCP server you want to add