Let your agent recon the web
One remote server, three lines of config. Your agent gets scrape, crawl, brand, screenshot, classify and PDF tools backed by the same API you call directly.
Status: the MCP server ships with the public launch. These instructions are final, and the endpoint goes live when it does. Get an API key now and it will work the moment the server is up.
- Transport
- Streamable HTTP
- URL
https://mcp.agentsrecon.com/mcp- Auth
Authorization: Bearer agr_...(your API key)
Run this once in your terminal. The server is saved to your user config.
claude mcp add --transport http agentsrecon https://mcp.agentsrecon.com/mcp \
--header "Authorization: Bearer agr_YOUR_KEY"Add this to claude_desktop_config.json, then restart Claude Desktop. The mcp-remote bridge speaks Streamable HTTP for you.
{
"mcpServers": {
"agentsrecon": {
"command": "npx",
"args": [
"-y",
"mcp-remote",
"https://mcp.agentsrecon.com/mcp",
"--header",
"Authorization: Bearer agr_YOUR_KEY"
]
}
}
}Add this to ~/.cursor/mcp.json, or to .cursor/mcp.json inside a project.
{
"mcpServers": {
"agentsrecon": {
"url": "https://mcp.agentsrecon.com/mcp",
"headers": {
"Authorization": "Bearer agr_YOUR_KEY"
}
}
}
}Tools your agent gets
- Scrape a page to clean Markdown
- Crawl a site, up to 5,000 pages, as an async job
- Extract brand: logo, colors, fonts
- Pull a full styleguide: CSS tokens and type scale
- Classify a business by NAICS and SIC
- Screenshot at several viewports, extract PDF text, search the web
Keep your key secret. Anyone with it spends your quota. Rotate it from the dashboard.
Prefer code over tools? The same calls ship as a TypeScript SDK and a Python SDK. Full reference lives in the docs.
Get a key and connect your agent
The Free plan includes the MCP server, rate-limited.