Connect MCP

Connect the research MCP

LinkDR exposes a Model Context Protocol server so Cursor, Grok, Claude, and other clients can call our research tools the same way they call Gmail or a database.

The server URL is:

{
  "mcpServers": {
    "linkdr": {
      "url": "https://linkdr.com/mcp"
    }
  }
}

Use HTTPS in production. The client should speak Streamable HTTP (POST). GET and DELETE are not used.

After you add the URL, the client discovers login from the 401 response. You should not put a bearer token in the JSON. See Auth.

What the tools can read

Once connected as an admin, it can:

  • Read a domain we already store (DR, traffic, keywords)
  • Search keywords we already tracked
  • Pull a stored Google SERP, and refresh it if the snapshot is older than 30 days
  • Search imported YouTube, Instagram, and TikTok creators
  • Look up contact emails we already have
  • Search the publisher marketplace

It cannot place an order, charge a card, send email, or invent a SERP we do not have.

The six tools

ToolTypical question
get_domainWhat do we know about this host?
search_keywordsWhich of our keywords match this phrase?
get_keyword_serpWho ranks for this keyword?
search_creatorsWhich creators match this name or topic?
get_contactsDo we have an email for this URL or domain?
search_domainsWhich publishers in the catalog match these filters?

Each tool returns JSON the model can read, plus a structured object for clients that prefer that.

Rate limit and empty answers

120 calls per user per hour. A tight keyword fan-out stays well under that. A loop over hundreds of keywords will not.

Empty answers are part of the contract. miss means the domain is not in the graph. no_snapshot means we could not return a SERP. The agent should say that out loud instead of filling gaps.

Pair MCP with a skill

Connecting MCP is not the same as knowing the workflow. Point the agent at keyword fan-out or YouTube outreach so it calls tools in a sensible order.