MCP servers give the agent tools Mindweave does not ship: issue trackers, databases, cloud APIs, internal services. Add one and it is available from that moment on, in every future session, with nothing to start by hand.
# a local server
/mcp add github npx -y @modelcontextprotocol/server-github --env GITHUB_TOKEN=ghp_x
# a remote one
/mcp add --http internal https://tools.acme.dev/mcp --header 'Authorization: Bearer t'
# available in every project, not just this one
/mcp add --global notes npx -y some-notes-server
# a server with its own flags: everything after -- goes to the server
/mcp add mine my-server -- --port 9000 --verbose
/mcp # what's running, and reconnect anything that isn't
/mcp remove x # stop configuring itYou can also just ask: "add the github mcp server, my token's in GITHUB_TOKEN." The agent writes the config and asks you to confirm before anything is saved.
Servers are declared in .mindweave/mcp.json (this project) or ~/.mindweave/mcp.json (everywhere), in the same format every other MCP client uses, so an existing config can be pasted straight in.
A changed tool description blocks the tool. If a server's tool description changes between sessions, that tool is held until you approve it. A changed description is the main way a server you already trusted turns hostile, so this is treated as a decision for you rather than a detail to absorb quietly.
OAuth is not supported yet. Remote servers that require it will show as needs-auth.
A server advertising many tools does not get to flood the model's tool list. Beyond a threshold its tools go into a deferred pool that the agent searches on demand, so a big catalog costs a search rather than a permanent slice of every request.
MCP is the youngest part of Mindweave and the part most likely to meet a server we have never seen. Resources and prompts have mostly been driven against servers written in-house, so real ones will find edges. Bug reports here are unusually valuable.