Introduction

The Graniite API gives you programmatic access to your library. There are two surfaces and both use the same per-user bearer token.

MCP server

The Model Context Protocol server is available at https://graniite.co/api/mcp. AI clients such as Claude Desktop, Cursor, ChatGPT, and n8n connect to it and call a small set of tools inside a conversation:

• search_knowledge_base: semantic, keyword, and hybrid search
• get_item: fetch a single item by id
• list_recent: list recent items in scope
• add_to_knowledge: add a URL or text snippet

The MCP surface is read and add only. Update, delete, and organization operations are available on the REST API.

REST API

The REST API is available at https://graniite.co/api/v1. It covers CRUD, organization, and automation, and is suitable for n8n flows, Zapier integrations, scripts, and custom code:

• Folders: list, create, delete
• Items: list, get, delete
• Memberships: add and remove items from folders
• Ingest: add URL or text content
• Uploads: three-step signed-URL flow for binary files

All endpoints are documented in the OpenAPI 3.1 spec. The spec works with any OpenAPI-aware tool wrapper, including LangChain OpenAPIToolkit, LlamaIndex OpenAPIToolSpec, and Postman.

Where to start

To connect an AI client to your library:

To build an automation that reads or writes your library programmatically:

To understand the model first: