FineData MCP Server
MCP (Model Context Protocol) server for FineData web scraping API.
Enables AI agents (Claude, Cursor, GPT, …) to fetch pages that need a full
browser and return clean data:
- JavaScript rendering and browser actions
- Captcha solving
- Datacenter / ISP / residential / mobile proxies
- Markdown or JSON output (markdown by default)
- AI structured extraction
- Local stdio or remote Streamable HTTP (+ OAuth 2.1)
Version: 0.3.1
Modes
Start with a plain request. Stealth and proxy modes are available when a page
needs a full browser; they consume more tokens than a plain request. Exact
rates are in the documentation and via get_usage. Gateway may apply a domain
strategy that overrides the requested engine or proxy; trust tokens_used in
the response.
Installation
uvx (recommended)
curl -LsSf https://astral.sh/uv/install.sh | sh
FINEDATA_API_KEY=fd_xxx uvx finedata-mcp
pip
pip install finedata-mcp
FINEDATA_API_KEY=fd_xxx finedata-mcp
npx
npx -y @finedata/mcp-server
Cursor
~/.cursor/mcp.json (Windows: %USERPROFILE%\.cursor\mcp.json):
{
"mcpServers": {
"finedata": {
"command": "uvx",
"args": ["finedata-mcp"],
"env": {
"FINEDATA_API_KEY": "fd_your_api_key_here"
}
}
}
}
Cursor deeplink (after publishing): install via MCP directory / “Add to Cursor”.
Remote (Streamable HTTP)
{
"mcpServers": {
"finedata": {
"url": "https://mcp.finedata.ai/mcp",
"headers": {
"Authorization": "Bearer fd_your_api_key_here"
}
}
}
}
OAuth 2.1 (Claude.ai / ChatGPT connectors): register via AS at https://api.finedata.ai, consent in the cabinet, then use the access token as Bearer.
Environment
| Variable | Required | Default | Description |
|---|
FINEDATA_API_KEY | stdio: yes | — | API key |
FINEDATA_API_URL | no | https://api.finedata.ai | API base |
FINEDATA_TIMEOUT | no | 180 | HTTP client timeout (seconds) |
FINEDATA_MCP_HOST | HTTP | 0.0.0.0 | Bind host |
FINEDATA_MCP_PORT | HTTP | 8080 | Bind port |
FINEDATA_OAUTH_ISSUER | remote OAuth | — | AS URL (gateway) |
FINEDATA_MCP_RESOURCE_URL | remote OAuth | — | Public MCP URL |
FINEDATA_JWT_SECRET | remote OAuth | — | Verify aud=mcp JWTs |
| Tool | Purpose |
|---|
scrape_url | Sync scrape with GET, read-only (markdown default) |
send_http_request | Sync POST / PUT / PATCH / DELETE through the same pipeline |
scrape_async | Async job, GET (formats=['markdown'] default) |
get_job_status | Poll job (markdown, not raw HTML) |
cancel_job | Cancel job |
list_jobs | List jobs |
batch_scrape | Up to 100 URLs (string or {url,...} objects) |
get_batch_status | Batch progress |
get_usage | Period usage via api_tokens_used |
Notable parameters: use_antibot, proxy_country, proxy_sticky, proxy_profile_id, auto_retry, stealth modes, formats, only_main_content, extract_*.
Async/batch: no csv/xlsx formats (sync only).
HTTP transport
finedata-mcp --transport http --host 0.0.0.0 --port 8080
Health: GET /health. Protected resource metadata is served at the path-scoped
URL for the endpoint — GET /.well-known/oauth-protected-resource/mcp — because
FINEDATA_MCP_RESOURCE_URL carries the /mcp path. Clients normally find it
from the WWW-Authenticate header on a 401 rather than guessing.
Support
License
MIT