MCP server (planned)
Planned — an MCP server so an AI assistant can read public social data directly. Not shipped; this page describes the intended design only.
Not shipped yet. There is no
@sourcevine/mcppackage to install and no hosted endpoint. This page records the intended design so the contract is settled before it is built — do not integrate against it.
What it will be
An MCP server, so an assistant that speaks the Model Context Protocol can resolve URLs and read public metrics itself. You give it a key; it gets the same four operations you would call from code, with the same credit costs and the same rate limits.
Planned naming: package @sourcevine/mcp, hosted endpoint
https://mcp.sourcevine.io/mcp.
Planned tools
| Tool | Maps to |
|---|---|
resolve_url |
POST /v1/resolve |
get_profile |
GET /v1/profile |
list_profile_posts |
GET /v1/profile/posts |
get_post |
GET /v1/post |
Responses are the same normalized objects documented in
the schema, including availability and the freshness fields.
Things that will be worth knowing
Credits are spent the same way. An assistant calling get_profile in a loop
costs exactly what your code would. Cache hits are still free, which makes
repeated questions about the same creator cheap.
Set a balance alert. An assistant deciding to check two hundred profiles is a normal thing for an assistant to do. An alert or an automatic top-up ceiling is worth having before you find out.
Use a separate key. Give the MCP server its own key on its own project, so you can see what the assistant spent and revoke it independently.
It cannot act on any platform. The tools read public data. There is nothing in the protocol surface that posts, follows, messages or changes anything anywhere.
Last updated 27 August 2026