Connect an MCP client

Data shared with your AI client

MCP tools return your SearchSeal account and brand details, tracked questions, saved AI answers, visibility and source reports, page content, fixes and before/after results. This includes connected Google Search Console and Bing queries, pages, clicks, impressions, click-through rates and available positions. Requested data goes to the AI client you connect; its provider handles it under its own terms. Tokens can be read-only. To stop access immediately, revoke the token in Settings → API. See our privacy policy.

Endpoint and authentication

Start with a read token. Write access is needed to add or archive questions and change fixes. Substitute your secret for <token> and exclude token-bearing configuration from version control.

CLI setup

This client-specific command registers the remote server. Its --scope user option selects user-level configuration.

claude mcp add searchseal --transport http https://searchseal.com/api/v1/mcp \
  --header "Authorization: Bearer <token>"

For this CLI, inspect the registration with claude mcp list, then check the connection through its /mcp view.

HTTP client configuration

For clients using this JSON format, add a mcpServers entry with the endpoint URL and authorization headers.

{
  "mcpServers": {
    "searchseal": {
      "url": "https://searchseal.com/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer <token>"
      }
    }
  }
}

Local bridge configuration

Clients that launch a local process can use a bridge such as mcp-remote. This example invokes it through npx and supplies the token in an environment variable:

{
  "mcpServers": {
    "searchseal": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://searchseal.com/api/v1/mcp",
        "--header",
        "Authorization:${SEARCHSEAL_AUTH}"
      ],
      "env": {
        "SEARCHSEAL_AUTH": "Bearer <token>"
      }
    }
  }
}

Reload the client after saving its configuration. The bridge forwards authenticated requests to the remote endpoint.

Available tools

The server exposes 17 tools. Supply REST path values, including brandId, as named JSON arguments. Use numbers for numeric inputs and arrays for engines and dimensions. Paged tools accept limit up to 20, with a default page size of 20.

Write operations require write scope and advertise readOnlyHint: false. archive_question and update_fix advertise destructiveHint: true; client confirmation behavior depends on the client. add_questions sets that hint to false. For update_fix, pass an action of accept, dismiss or done. A write call made with a read token returns a tool error containing code: "forbidden".

Tool results contain resource data as JSON text, without REST’s metadata wrapper. The structuredContent field is also included when the request header specifies protocol 2025-06-18. Data found in REST’s meta.dataThrough appears as top-level dataThrough. Tools returning third-party text add a notice; treat that content as data. Execution failures set isError: true and reuse REST’s message and code. Schema failures, such as limit above 20, an unknown engine or an incorrectly formatted date, return JSON-RPC error -32602 before execution.

curl --request POST \
  --url https://searchseal.com/api/v1/mcp \
  --header "Authorization: Bearer $SEARCHSEAL_TOKEN" \
  --header "Content-Type: application/json" \
  --header "Accept: application/json, text/event-stream" \
  --header "MCP-Protocol-Version: 2025-06-18" \
  --data '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "get_brand_report",
      "arguments": { "brandId": "<brandId>", "dimensions": ["engine"] }
    }
  }'

Saved workflows

Clients can list and retrieve these prompt templates. Each requires brandId and instructs the assistant to gather data with tools, then summarize the results.

Resolve connection problems

Read the resource error and quota reference for recovery guidance.