WebGlean

The webglean-mcp server exposes the WebGlean API to any MCP-compatible host — Claude Desktop, Claude Code, or your own agent — as a set of callable tools. It's built on top of the Node.js SDK and talks to the host over stdio.

Setup

Add it to your MCP host's config with your API key in env:

{
  "mcpServers": {
    "webglean": {
      "command": "npx",
      "args": ["-y", "webglean-mcp"],
      "env": {
        "WEBGLEAN_API_KEY": "wg_your_api_key"
      }
    }
  }
}

For Claude Code, use the CLI instead of editing JSON directly:

claude mcp add webglean --env WEBGLEAN_API_KEY=wg_your_api_key -- npx -y webglean-mcp

WEBGLEAN_BASE_URL can also be set in env to point at a non-default API host. There's no interactive login for this server — unlike the CLI, auth is env-only, since the host application launches the process and sets env itself.

Package page: webglean-mcp on npm.

Tools

Tool	Endpoint
`webglean_scrape`	`POST /v1/scrape`
`webglean_crawl` / `webglean_crawl_status`	`POST /v1/crawl`, `GET /v1/crawl/:id`
`webglean_extract`	`POST /v1/extract`
`webglean_map`	`POST /v1/map`
`webglean_search`	`POST /v1/search`
`webglean_monitor_create` / `webglean_monitor_list` / `webglean_monitor_get` / `webglean_monitor_delete`	`POST` / `GET` / `GET` / `DELETE /v1/monitor`
`webglean_batch_scrape` / `webglean_batch_status`	`POST /v1/batch/scrape`, `GET /v1/batch/scrape/:id`

Behavior worth knowing

webglean_crawl and webglean_batch_scrape wait for the job to finish by default — pass wait: false to get a job id back immediately and poll it yourself with webglean_crawl_status / webglean_batch_status instead.
Both of those tools also trim their response down to url + markdown per page/result, dropping html, text, and metadata to keep tool output compact for the calling agent's context window. Call webglean_scrape directly on a specific URL if you need its HTML, plain text, or metadata.
Tool-level failures come back as a normal result with isError: true and the API's error message (plus HTTP status, when there is one) — not a thrown protocol-level error — so the calling agent can read the failure and decide how to react, rather than the whole tool call blowing up.
webglean_batch_scrape normalizes bare-domain URLs (example.com → https://example.com) before sending, since that's the one API endpoint that doesn't do this automatically server-side.

Trying it interactively

npx @modelcontextprotocol/inspector npx -y webglean-mcp

This opens the MCP Inspector so you can call each tool by hand and see the raw request/response before wiring it into an agent.