MCP servers

MCP is an open protocol for exposing tools and data to AI assistants. Open Interpreter can connect to any MCP server and use its tools alongside the built-in ones.

Common uses:

• Pull issues from a tracker
• Query a private knowledge base
• Run domain-specific commands the agent should not invent

Configure a server

Add it to ~/.openinterpreter/config.toml under [mcp_servers.<name>]:

[mcp_servers.linear]
command = "npx"
args = ["-y", "@linear/mcp-server"]
env = { LINEAR_API_KEY = "lin_api_..." }

The server is launched on demand and shut down when you exit the session.

Browse what is loaded

Inside a session:

/mcp

Add verbose to see every tool a server exposes:

/mcp verbose

Approval modes

By default, every MCP tool call asks for approval. Loosen it for trusted servers:

[mcp_servers.docs]
command = "docs-server"
default_tools_approval_mode = "approve"

Override individual tools:

[mcp_servers.docs.tools.delete_doc]
approval_mode = "prompt"

Parallel tool calls

MCP tools run one at a time by default. If a server's tools are safe to run together, opt in:

[mcp_servers.search]
command = "search-mcp"
supports_parallel_tool_calls = true

Remote servers

Open Interpreter can also connect to remote MCP servers over a URL.

[mcp_servers.acme]
url = "https://mcp.acme.com"
auth_token_env = "ACME_TOKEN"

The token is read from the named environment variable so secrets stay out of the file.

Troubleshooting