MCP setup

@crawlpipe/mcp is a stdio MCP server that wraps the same /v1/legal/* endpoints documented in the API reference — every tool call is a real, metered REST request under the hood, so usage shows up in Usage exactly like direct API calls.

Run it

npx @crawlpipe/mcp

No install step needed for most MCP clients — they invoke npx on demand.

Configuration

The server reads two environment variables:

Variable	Required	Meaning
`CRAWLPIPE_API_KEY`	Yes	An API key from API Keys in the portal (`dp_live_…`)
`CRAWLPIPE_BASE_URL`	No	Overrides the API base URL (defaults to `https://api.crawlpipe.com`) — useful for local development against `http://localhost:8000`

Example: Claude Desktop / Claude Code config

{
  "mcpServers": {
    "crawlpipe": {
      "command": "npx",
      "args": ["-y", "@crawlpipe/mcp"],
      "env": {
        "CRAWLPIPE_API_KEY": "dp_live_a1b2c3d4e5f6…"
      }
    }
  }
}

Tools

Tool	REST mapping
`legal_search(q, mode?, filters?)`	`GET /v1/legal/search`
`legal_get_document(id, at?, include?)`	`GET /v1/legal/documents/{id}`
`legal_get_effectivity(id, at?)`	`GET /v1/legal/documents/{id}/effectivity`
`legal_get_relations(id, direction?, types?)`	`GET /v1/legal/documents/{id}/relations`
`legal_changes(since, filters?)`	`GET /v1/legal/changes`
`crawlpipe_usage()`	`GET /v1/usage`

Tool descriptions are written for agent consumption — they include the vocab enums (doc_type, eff_status, relation_type, …) and the same citation guidance as RAG with the search API, so an agent calling legal_search gets a ready-to-cite result without extra prompting.

Metering

Because the MCP server calls the same REST API with your key, every tool invocation is metered exactly like the equivalent REST call — legal_search with mode: "hybrid" costs 2 units through MCP too. There's no separate MCP pricing or quota.