CrispParse MCP Server
Official# CrispParse MCP Server
A self-contained Model Context Protocol (`stdio`) server implementation built with `@modelcontextprotocol/sdk` that exposes deterministic web extraction tools to MCP-compatible AI agents and desktop clients (Claude Desktop, Cursor, Windsurf, and Cline).
## How It Works
This repository implements a local MCP `stdio` server (`index.js`) using `StdioServerTransport`:
1. Registers MCP `tools/list` and `tools/call` handlers via `@modelcontextprotocol/sdk`.
2. Exposes the `crispparse_extract` tool with JSON Schema input validation.
3. Executes upstream extraction requests against the CrispParse extraction engine (`https://api.crispparse.com/v1/extract`) and returns clean Markdown or JSON over `stdio`.
## Available Tools
### `crispparse_extract`
Extracts clean, LLM-ready Markdown or structured JSON from any public webpage URL.
* **`url`** (`string`, required): The full `http://` or `https://` URL to extract.
* **`mode`** (`string`, optional, default `"markdown"`): Output format (`"markdown"` or `"json"`).
## Installation & Local Execution
### Run with Node.js (v18+)
git clone https://github.com/goldmarklabs/crispparse-mcp.git
cd crispparse-mcp
npm install
CRISPPARSE_API_KEY="your_api_key" node index.js
### Run with Docker
docker build -t crispparse-mcp .
docker run -i --rm -e CRISPPARSE_API_KEY="your_api_key" crispparse-mcp
## MCP Client Configuration
Add the server to your Claude Desktop (`claude_desktop_config.json`) or Cursor MCP configuration:
{
"mcpServers": {
"crispparse": {
"command": "node",
"args": ["/absolute/path/to/crispparse-mcp/index.js"],
"env": {
"CRISPPARSE_API_KEY": "your_api_key"
}
}
}
}
## Environment Variables
* `CRISPPARSE_API_KEY`: Optional Bearer Macaroon API key from [crispparse.com](https://crispparse.com) (requests without a key receive an `x402` Base USDC payment challenge).
* `CRISPPARSE_API_URL`: Optional upstream endpoint override (defaults to `https://api.crispparse.com/v1/extract`).
## License
MIT — see [LICENSE](./LICENSE).
TDQS
Scored across 1 tool
With only a single tool in the server, there is no possibility of confusion or misselection between tools. The purpose of crispparse_extract is unambiguous.
The lone tool uses a clear, namespaced snake_case verb_noun pattern (crispparse_extract) that matches the server name. There is no competing convention to conflict with it.
A single tool is defensible for a narrow URL-to-Markdown extraction service, but it is thin: there is no companion tool for format discovery, batch extraction, or error/usage inspection. The surface feels minimal even for its focused scope.
For the core domain of fetching and converting a public URL into Markdown or structured JSON, one tool covers the primary operation end to end. Minor gaps remain around batch processing and explicit output-format/schema configuration, but agents can work around these via parameters.