docAnalyzer

Connect with MCP

Point Claude Desktop, Cursor, or any MCP client at your documents using the MCP server docAnalyzer publishes.

docAnalyzer publishes a ready-to-run MCP server. You don't need to write one. It exposes each REST API operation as an MCP tool and forwards tool calls to the API using your API key, so an MCP client can list your documents, upload new ones, and chat with them.

Two files are published next to the API reference:

File What it is
mcp-server.mjs A runnable MCP server using the stdio transport. This is the file your client launches.
mcp.json A manifest describing the tools and the auth shape. Useful for inspection or for generating your own client.

mcp.json is a manifest, not a server address. Putting its URL in the url field of a client config won't work, because nothing at that address speaks MCP. docAnalyzer doesn't host a remote MCP endpoint today; the server runs on your machine as a local process and talks to the API over HTTPS.

Before you start

Set up

  1. 1

    Download the server into a directory of your choice:

    curl -O https://docanalyzer.ai/docs/api-reference/mcp-server.mjs
  2. 2

    Install the MCP SDK in that same directory:

    npm i @modelcontextprotocol/sdk
  3. 3

    Register the server with your MCP client. Most clients (Claude Desktop, Cursor, and others) use a config file with this shape:

    {
  "mcpServers": {
    "docanalyzer": {
      "command": "node",
      "args": ["/absolute/path/to/mcp-server.mjs"],
      "env": {
        "API_TOKEN": "your-api-key"
      }
    }
  }
}

    Use an absolute path. Clients don't launch the server from the directory you downloaded it into.

  4. 4

    Restart the client, then confirm the docAnalyzer tools appear in its tool list.

Environment variables

Variable Required Purpose
API_TOKEN Yes Your API key. Sent as Authorization: Bearer <key> on every call.
API_BASE_URL No Overrides the API base. Defaults to https://api.docanalyzer.ai/api/v1.

Keep the key in the client's env block or your operating system's keychain. Don't paste it into mcp-server.mjs.

Available tools

The tools are generated from the API reference, one per operation, so the list tracks the API:

Tool Does
get_doc List all documents in a workspace
post_doc Create a document from a URL
post_doc_upload_options Upload documents
get_doc_docid Get document information
put_doc_docid Update a document
delete_doc_docid Delete a document
get_doc_docid_link Get a link to a document
post_doc_docid_chat Chat with a document
post_doc_docid_ocr Run OCR on a document
get_label, post_label, get_label_labelid, put_label_labelid, delete_label_labelid Manage labels
post_label_labelid_upload_options Upload documents into a label
post_label_labelid_chat Chat with every document in a label
get_models, get_models_options List available models
get_credits Read credit usage for the key

Chat tools start a new conversation thread on each call. To continue a conversation, capture the tid from the response and pass it back as tid. Only one chat can be in flight per tid; a concurrent call on the same tid returns 423.

The server covers the v1 API. Agentic Workflows (v2) aren't exposed as MCP tools; call that API directly.

Quotas and credits

Tool calls are ordinary API calls. They count against the same uploads, storage, interaction, and credit limits as the app, and chat tools spend credits the same way a chat in the app does. See Plans and credits.

Other agent files

The same directory publishes two plain-text descriptions of the API for agents that read documentation rather than call tools:

What's next

Was this helpful?