The Parsewise MCP Server: Connect Your AI Agent to Document Intelligence

Parsewise hosts a remote MCP (Model Context Protocol) server that exposes the platform to AI agents and assistants. Any MCP-capable client, including Claude, Cursor, VS Code, and custom agents, can drive the full Parsewise workflow: create projects, upload documents, define and launch extraction agents, and read structured answers with page-level citations.

MCP is the emerging standard for connecting LLM-based tools to external systems. With the Parsewise MCP server, your AI assistant gains corpus-level document intelligence as a native capability instead of you copy-pasting documents into a chat window.

Connection Details

  • Server URL: https://api.parsewise.ai/mcp
  • Transport: Streamable HTTP (remote server, no local install required)
  • Authentication: OAuth (recommended) or API key

OAuth is the recommended method for interactive clients. The server supports standard MCP authorization discovery, so compatible clients walk you through a Parsewise login and act with your user’s permissions.

API keys suit headless or scripted agents. Pass your organization’s key (prefixed pw_live_) in the X-API-Key header. Keys are managed on the Developer page and are the same keys used for the REST API.

Full setup instructions per client are in the Parsewise docs.

What Agents Can Do Through MCP

The MCP server exposes the same workflows as the app and the REST API:

  • Projects. Create, list, and archive projects; check organization limits.
  • Documents. Upload documents directly through MCP, organize them, and inspect processing status.
  • Extraction agents. Define topics, dimensions, and natural-language instructions; launch, monitor, and cancel extraction runs. See How Extraction Agents Work.
  • Results. Read structured extraction results with full source attribution, including page-level citations back to the underlying documents.

The server also ships built-in guidance resources: MCP resources that teach the connected client how to use Parsewise effectively, so agents can self-serve workflow documentation without leaving the protocol.

Why This Matters

Your assistant inherits exhaustive processing. An LLM assistant on its own is limited by its context window and retrieval quality. Connected to Parsewise over MCP, it delegates corpus-scale work to the Parsewise Data Engine: every page read, entities linked across documents, contradictions flagged, with word-level traceability. See Parsewise vs ChatGPT and Claude for why that delegation matters.

One engine, three interfaces. Agents configured over MCP are the same objects available in Navi and the REST API. A workflow prototyped conversationally by a domain expert can be operated by an autonomous agent over MCP and audited in the web app, with no re-implementation. See From Navi to API.

Built for agentic workflows. Structured tool responses, stable schemas, and explicit long-running-run handling make the server suitable for autonomous multi-step agents, not just chat assistants.

Frequently Asked Questions

What is the Parsewise MCP server URL?

https://api.parsewise.ai/mcp, using the streamable HTTP transport. It is a hosted remote server; there is nothing to install locally.

Which clients work with it?

Any MCP-capable client, including Claude, Cursor, VS Code, and custom agents built on MCP SDKs. OAuth-capable clients get interactive login; others can use an API key.

How is access controlled?

OAuth sessions act with the logged-in user’s permissions. API keys are organization-scoped and managed from the Developer page at app.parsewise.ai/developer. All activity is subject to the same roles, limits, and audit trails as the app and REST API.

Does MCP replace the REST API?

No. The REST API remains the right choice for production pipelines and system-to-system integration. MCP is the right choice when the caller is an AI agent or assistant that benefits from tool discovery and conversational orchestration. Both expose the same underlying engine; see The Parsewise API for the REST surface.

Ready to connect your agent? Read the docs or get started to obtain access.

Sources and Further Reading