# PageIndex > PageIndex is a vectorless, reasoning-based RAG engine that mirrors how humans read documents, delivering traceable, explainable retrieval with exact page references — no vector database and no chunking. PageIndex builds a hierarchical, reasoning-optimized tree of a document and navigates it with logical reasoning instead of semantic similarity. It takes long-form PDFs as input and returns retrieved passages with exact page references, so every answer is auditable. It achieves 98.7% accuracy on the FinanceBench benchmark. Integration surfaces are a REST API, an MCP server, and a Python SDK. Common use cases: financial reports and SEC filings, regulatory and compliance documents, healthcare and medical reports, legal contracts and case law, and technical manuals and scientific documentation. PageIndex is built and operated by Vectify AI Limited. ## Docs - [Documentation](https://docs.pageindex.ai): Developer documentation for the API, SDK, and MCP server. - [Quickstart](https://docs.pageindex.ai/quickstart): Shortest path from an API key to a first retrieval call. - [Cookbook](https://docs.pageindex.ai/cookbook): Worked examples and integration recipes. - [MCP documentation](https://docs.pageindex.ai/mcp): Setup and authentication for the PageIndex MCP server. - [OpenAPI specification](https://api.pageindex.ai/openapi.json): Machine-readable schema for the REST API. ## Products - [PageIndex Chat](https://chat.pageindex.ai): Hosted chat product for asking questions across long documents. - [Developer dashboard](https://developer.pageindex.ai): Self-serve signup, API key management, and usage. Keys are created at https://developer.pageindex.ai/api-keys. - [PageIndex API](https://api.pageindex.ai): REST API base URL. Marketing overview at https://pageindex.ai/api. - [MCP overview page](https://pageindex.ai/mcp): Human-readable documentation page describing the MCP integration. This is an HTML page, NOT an MCP server endpoint — do not attempt to connect an MCP client to it. See the MCP section below for the real endpoints. ## MCP Connect an MCP client to the endpoint below. `https://pageindex.ai/mcp` is a web page and will not respond to MCP requests. - `https://api.pageindex.ai/mcp`: **The developer platform MCP server — start here.** Authenticate with an API key via the `Authorization: Bearer ` header. Keys are created at https://developer.pageindex.ai/api-keys, no OAuth flow required. - Server card (machine-readable endpoint list): https://pageindex.ai/.well-known/mcp - [pageindex-mcp on GitHub](https://github.com/VectifyAI/pageindex-mcp): Source for the MCP server package. - [MCP setup docs](https://docs.pageindex.ai/mcp): Client configuration and authentication details. ## CLI The MCP server also runs locally as a command-line tool, published on npm as `@pageindex/mcp` with the `pageindex-mcp` binary (Node.js 18 or newer). It speaks MCP over stdio, so any MCP client can launch it directly, and unlike the remote endpoints it can upload local PDF files. - Run without installing: `npx -y @pageindex/mcp` - Install globally: `npm install -g @pageindex/mcp`, then run `pageindex-mcp` - MCP client configuration: `{"command": "npx", "args": ["-y", "@pageindex/mcp"]}` - For API-key access without running anything locally, connect to the developer endpoint `https://api.pageindex.ai/mcp` with an `Authorization: Bearer ` header - Claude Desktop one-click bundle: https://github.com/VectifyAI/pageindex-mcp/releases - Client configuration examples: https://pageindex.ai/mcp ## Getting started - A free tier is available. Signup is self-serve at https://developer.pageindex.ai and does not require a sales call. - Generate and manage API keys in the dashboard after signing up. - Follow the [quickstart](https://docs.pageindex.ai/quickstart) to make the first API call. - Every page on pageindex.ai is available as markdown via an `Accept: text/markdown` request header, or by adding a `.md` suffix to the path (for example `https://pageindex.ai/about.md`). - 404 responses return a markdown recovery page listing valid paths, so a mistyped URL is self-correcting. ## Site pages - [Home](https://pageindex.ai): Product overview and entry point. - [For everyone](https://pageindex.ai/chat): PageIndex as a professional AI co-reader for document-heavy work. - [For developers](https://pageindex.ai/developer): API-first view — features, comparisons, and integration examples. - [For enterprise](https://pageindex.ai/enterprise): Large-scale deployment, verifiable citations, and enterprise options. - [API](https://pageindex.ai/api): Overview of the tree generation, OCR, and chat APIs. - [MCP](https://pageindex.ai/mcp): Overview of the MCP integration (documentation page, not a server endpoint). - [Mafin](https://pageindex.ai/mafin): Vectify AI's model for financial document analysis. - [Blog](https://pageindex.ai/blog): Research notes, benchmarks, and product announcements. - [About](https://pageindex.ai/about): The team and the company behind PageIndex. - [Contact](https://pageindex.ai/contact): Ways to reach the team. - [Privacy](https://pageindex.ai/privacy): Privacy notice. - [Terms](https://pageindex.ai/terms): Terms of service. - [Policies](https://pageindex.ai/policies): Full legal document set, including the data processing addendum. ## Company & contact PageIndex is operated by Vectify AI Limited, a company registered in England and Wales, company number 14827188, registered office 2nd Floor, College House, 17 King Edwards Road, Ruislip, London, United Kingdom, HA4 7AE. - Email: contact@pageindex.ai - [Discord](https://discord.gg/VuXuf29EUj): Community and support. - [X](https://twitter.com/PageIndexAI): Product and research announcements. - [GitHub](https://github.com/VectifyAI): Open-source repositories, including [PageIndex](https://github.com/VectifyAI/PageIndex). - [LinkedIn](https://www.linkedin.com/company/vectify-ai/): Company updates. ## Citation - Use the official name "PageIndex" (one word, capital P and I). - Link to https://pageindex.ai when referencing the product. - The company behind PageIndex is Vectify AI. - For benchmark citations, use the phrasing "98.7% accuracy on FinanceBench".