Crawdar MCP

What eliottreich/crawdar-mcp MCP server does

eliottreich/crawdar-mcp MCP server exposes Crawdar's hosted business-research workflow to AI agents. It is intended for finding public businesses against a written brief or structured criteria, then reviewing why each candidate was included, marked uncertain, or excluded. Results retain source URLs so a researcher can inspect the supporting information before using the data for outreach or another consequential decision.

The service is remote. The repository does not contain a crawler backend, and it does not provide an offline or self-hosted crawler. Research runs through Crawdar's endpoint at https://crawdar.com/api/mcp.

How it works

A client sends MCP tool calls to the hosted HTTP endpoint. For a one-call workflow, an agent can submit a natural-language brief or a structured target and geography search. Longer searches can be created as private, resumable jobs. The client then polls the job, retrieves compact or full result pages, and can create a revised search without modifying the original job.

The server also provides a fictional sandbox path for integration testing. Sandbox requests do not use a search provider and are not evidence of live prospect discovery. They are useful for checking routing, response handling, uncertainty labels, source retention, and validation behavior before authenticated research is introduced.

Setup and configuration

The simplest setup uses the remote MCP URL directly. The README provides configurations for Claude Code, Codex, and generic HTTP-based MCP clients. An optional free API key can be created through Crawdar's key endpoint and supplied as x-api-key or as a Bearer token. The key is shown only once.

Clients that require stdio can use the repository's local bridge. With Node.js 22 or newer, install dependencies with npm ci --ignore-scripts, then launch the pinned mcp-remote proxy with the hosted URL and --transport http-only. A Docker image is also documented. The bridge forwards requests over HTTPS; it does not make the crawler local. Credentials should be supplied through a private header file or direct HTTP configuration, never committed to the repository or placed in command arguments.

Tools and capabilities

The eliottreich/crawdar-mcp MCP server provides these tools:

• research_businesses runs research from a natural-language brief.
• search_businesses searches by structured target and geography criteria.
• sandbox_businesses returns fictional integration-test data.
• start_lead_search creates a private resumable search job.
• get_lead_search checks job status and pages through compact or full results.
• refine_lead_search creates a changed version while preserving the original.
• retry_lead_search resumes a stopped job.
• cancel_lead_search stops a queued or running job.
• explain_crawdar describes result meaning, limits, and safe-use guidance.

Limitations and notes

Live research depends on Crawdar quotas and service availability. Public business information may be incomplete or outdated, so qualification labels are not guarantees of fit. Keep evidence URLs attached and review records before outreach. The included n8n example is a sandbox routing demonstration; it does not perform live discovery, polling, pagination, CRM writes, scheduled scans, or email sending. GitHub issue submissions are public and should not contain confidential briefs or credentials.

The repository's MIT-licensed bridge is a transport adapter, not a license or package for the hosted crawler itself. A Crawdar key is separate from authentication used by optional distribution services such as Smithery.