Last updated: August 30, 2026

Limurse MCP server

Connect Limurse to Claude, ChatGPT or any other MCP client, and find bookable creators, benchmark rates and research competitor ads from inside the conversation.

Limurse publishes a remote Model Context Protocol server. Once connected, an assistant can search Limurse's creator marketplace on your behalf using your own account's access.

The server is read-only. It exposes no tool that creates, edits or sends anything: no collaboration, no invitation, no status change, no payment. The closest thing to an action is start_booking, which returns a link for you to open.

Connect it

Add this URL as a custom connector wherever your client accepts one:

https://api.limurse.ai/mcp

Your client will open a Limurse consent screen in the browser, where you sign in and approve the access it asked for. Nothing is shared until you approve it, and you can revoke a connection at any time from Settings → Connected apps on Limurse.

For a client configured from a file rather than a UI, the same URL goes in its MCP server list:

{
  "mcpServers": {
    "limurse": {
      "type": "http",
      "url": "https://api.limurse.ai/mcp"
    }
  }
}

Tools

search_creators

Search the directory of bookable creators by niche, location, audience size and the brands they have worked with. Returns public profile cards, never contact details or individual fees.

Requires the creators:read scope.

get_creator_profile

Read one creator’s full public profile by username: bio, content categories, brands worked with, portfolio media and reach.

Requires the creators:read scope.

creator_rate_benchmarks

Aggregate market rates by niche, for budgeting a campaign. Bands are computed across the marketplace; no individual creator’s rate card is exposed.

Requires the creators:read scope.

search_brand_ads

Search the Meta Ad Library for ads a brand is currently running on Facebook and Instagram. Public archive data, not the brand’s ad account.

Requires the ads:read scope.

start_booking

Return the Limurse link where a brand completes a booking. This books nothing. It hands back a URL for a human to open.

Requires the creators:read scope.

search_brand_ads depends on Meta Ad Library credentials and is only listed by a deployment that carries them. A client that does not see it is connected to one that does not.

Authentication

The server implements OAuth 2.1 with PKCE, and Limurse is the authorization server. It supports dynamic client registration, so most MCP clients can connect without anyone issuing them credentials by hand.

  • Protected resource metadata: https://api.limurse.ai/.well-known/oauth-protected-resource
  • Authorization server metadata: https://api.limurse.ai/.well-known/oauth-authorization-server
  • Scopes: creators:read for the creator tools, ads:read for ad research. Both are read-only; there is no write scope.
  • Transport: Streamable HTTP on POST /mcp, stateless. The server opens no event stream, so GET /mcp answers 405 by design.
  • Protocol revision: 2025-06-18.

Access tokens are short-lived and refresh tokens rotate. Re-using a rotated refresh token revokes the whole chain, so a leaked token stops working rather than quietly persisting.

What the connector can and cannot see

A connected assistant sees exactly what an anonymous visitor to Limurse's public creator search sees: the same records, the same redactions. It does not inherit your workspace's private data.

  • Never exposed: creator email addresses, phone numbers, postal addresses, individual rate cards, collaboration preferences, or any other workspace's data.
  • Never written: no tool in this server modifies anything on Limurse.
  • Rates are aggregates only. creator_rate_benchmarks returns bands across the marketplace, not what any one creator charges.

Data handling is governed by our Privacy Policy, and use of the connector by our Terms and Conditions.

Support

Report a problem with the connector, or ask for access to something it does not yet expose, at apps@limurse.ai. For a suspected security issue, follow our Security Policy instead.