Integrations & API

MCP server

Let Claude, Cursor or any MCP client call Rixe as a tool to read your scores, gaps, competitors, brand kit and knowledge base.

The Model Context Protocol (MCP) is an open standard that lets an AI assistant call outside tools. Rixe runs an MCP server, so an assistant such as Claude Desktop or Cursor can answer questions like "which prompts are we invisible in on Perplexity?" from your live Rixe data, right where you are already working, without opening the dashboard.

The server uses the same API keys and scopes as the REST API. Each tool asks for its own scope, so a key can be limited to exactly what the assistant should see.

Where to find it

Open the user menu at the bottom of the sidebar and go to Settings › Developer. The MCP server card shows the Server URL, a Client config (Claude Desktop / Cursor) snippet, a Quick test command and the list of tools, each with a Copy button.

Server URLhttps://api.rixe.ai/mcp
TransportJSON-RPC 2.0 over a single HTTP POST
Protocol version2025-06-18
AuthenticationAuthorization: Bearer rixe_… or x-api-key: rixe_…

MCP access comes with the "REST API + MCP" feature: you need a plan that can create API keys. See Plans and limits.

Set it up

  1. In Settings › Developer › API keys, click Add API key, name it (for example "Claude Desktop") and tick the scopes for the tools you want (see the table below). Click Generate key and copy it; it is shown only once.
  2. Copy the Client config snippet from the MCP server card into your client's MCP configuration. For Claude Desktop that is claude_desktop_config.json; for Cursor, its MCP settings file.
  3. Replace rixe_… with your key and restart the client.

The snippet uses the mcp-remote bridge, which runs through npx (Node.js must be installed):

json
{
  "mcpServers": {
    "rixe": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://api.rixe.ai/mcp",
        "--header",
        "Authorization: Bearer ${RIXE_API_KEY}"
      ],
      "env": { "RIXE_API_KEY": "rixe_YOUR_KEY" }
    }
  }
}
  1. Ask the assistant something like "Using Rixe, list my workspaces and summarise visibility for the first one." It should call calisma_alanlari first, then gorunurluk_ozeti.

Quick test from the command line

bash
curl -s https://api.rixe.ai/mcp \
  -H 'Authorization: Bearer rixe_YOUR_KEY' -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'

A working key returns {"jsonrpc":"2.0","id":1,"result":{"tools":[…]}}.

Tools

Tool names are fixed identifiers (they happen to be Turkish words); your assistant reads the English description of each and picks the right one. Every tool except calisma_alanlari needs a workspaceId. Where days is accepted, the default window is 30 days and the maximum is 180.

ToolScopeArgumentsReturns
calisma_alanlarinonenoneThe workspaces the key can read, with id and name. Start here.
gorunurluk_ozetiread:scoresworkspaceId, daysThe composite visibility score, its confidence interval, the trend and a card per engine. Rates with too few measurements are reported as insufficient, not as numbers.
rakiplerread:scoresworkspaceId, daysBrand-by-brand comparison: mention rate, average position, citation rate, sentiment and rank (with the previous rank).
soru_tablosuread:promptsworkspaceId, daysPass rate per question, engine by engine, with the change since the previous period.
bosluklarread:promptsworkspaceId, daysPrompts where a competitor is recommended and you are not, most important first.
kapsamread:promptsworkspaceIdHow your tracked questions spread over the buyer journey (awareness, consideration, evaluation, purchase) and how many to add.
rakip_kaynaklariread:sourcesworkspaceId, daysDomains cited in answers that named a rival and not you, with the rivals and questions behind each.
yapilacaklarread:actionsworkspaceIdThe top 20 open actions, ranked by impact × confidence / effort.
marka_kitiread:brandworkspaceIdThe brand's writing context: voice, writing rules, claims to make and avoid, author persona.
bilgi_tabaniread:knowledgeworkspaceId, query, limit (default 6, max 20)Knowledge-base passages that match the query, best first.
asistanask:assistantworkspaceId, question, locale (default en)An answer from Rixe's assistant, based on the workspace's current numbers and knowledge base, plus its cost.

`asistan` costs money

Each asistan call is one model call billed to your organization, and the response includes costUsd. It needs the ask:assistant scope, which a key never gets unless you tick it explicitly. See Assistant.

A read-only analyst key

For most people read:scores, read:prompts, read:sources and read:actions are enough to ask "where are we losing and why". Add read:brand and read:knowledge if the assistant will draft content in your name.

Calling a tool directly

The server answers the MCP methods initialize, ping, tools/list and tools/call. A tool result is a single text block containing JSON:

bash
curl -s https://api.rixe.ai/mcp \
  -H 'Authorization: Bearer rixe_YOUR_KEY' -H 'Content-Type: application/json' \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"calisma_alanlari","arguments":{}}}'
json
{
  "jsonrpc": "2.0",
  "id": 2,
  "result": {
    "content": [
      {
        "type": "text",
        "text": "[\n  {\n    \"id\": \"9b1f6c2e-4d1a-4f0e-8a51-2f3c7d9e1a10\",\n    \"name\": \"FlowCRM – US\"\n  }\n]"
      }
    ]
  }
}

Errors

  • No key or a revoked key: the HTTP request itself fails with 401, before any JSON-RPC reply.
  • Missing scope, unknown tool or missing argument: the HTTP status is 200 and the reply carries a JSON-RPC error with code -32603, for example This API key is missing the "read:actions" scope or query is required. Tools the key can use keep working; only the one call fails.

Limits and good to know

  • The server is stateless: there are no sessions, streaming or server-initiated messages. Each call is a single request and response.
  • It shares the API's rate limit of 1,200 requests per minute.
  • The data is the same as in the app: scores follow the daily measurement, and a rate built on fewer than 4 measurements is marked as not enough data rather than given as a number.
  • MCP is read-only apart from asistan, which spends money but does not change your data.

Still stuck? Write to [email protected] and a person will answer.