MCP servers
The Model Context Protocol (MCP) is an open, transport-agnostic protocol for connecting an LLM to an external server that exposes tools, resources, and prompts. It lets a product point a model at an arbitrary third-party service without hard-coding a client for each one.
Not supported yet
Vibesboard has no MCP client, no mcp:* tool type, and no place to configure
an MCP server — in the UI or the API. An agent cannot be connected to one
today. This page exists so the gap is documented rather than discovered.
What exists instead
Each of these is narrower than MCP: a fixed set of tools defined by the platform, rather than a way to point an agent at an arbitrary tool server.
- Data actions & tools — an agent can append or update records in Google Sheets, Airtable, or a webhook you control. The tool shapes (
submit_data,update_record) are fixed by the platform, not by the destination. - Webhooks — the most generic data-action provider: Vibesboard
POSTs orPUTs a fixed JSON envelope to a URL you configure. You control what happens on your side, but the request and response contract is set here, not negotiated per tool the way MCP does it. builtin:web_fetch— lets an agent fetch and read the text of an arbitrary URL. A one-way read, not a callable tool server.- Hooks & lifecycle — the inverse direction: an external system invokes an agent and receives an HMAC-signed callback.
If you specifically need an agent to discover and call tools on an arbitrary external server, none of these do it. Plan around the gap rather than around a roadmap.