About MCP

MCP (the Model Context Protocol) is a new open standard that major AI companies (Anthropic, OpenAI, and others) have adopted to let tools like Engram plug directly into AI assistants. Think of it as a universal connector: one standard, and Claude, Cursor, ChatGPT, or any other AI tool that supports MCP can reach your notes. They can search them, read them, and save new ones, all with permissions you control.

You don’t need to understand MCP to use Engram. Pick your AI tool in AI Integrations, follow a few clicks, and you’re done. This page is here if you want the background.

What MCP is

Before MCP, getting your notes into an AI assistant meant copy-pasting, exporting, or building a custom integration per app. MCP standardizes the connection:

• Engram exposes a single MCP endpoint (one URL, one protocol).
• Your AI client speaks MCP (Claude, Cursor, ChatGPT, VS Code, Windsurf, Zed, and more).
• The handshake is OAuth. You stay in control of what’s reachable through which client.
• Tools, prompts, and resources are discoverable. The client asks Engram what’s available; Engram answers. Nothing pre-baked per integration.

A useful one-line metaphor: MCP is USB-C for AI. One wire format, many devices on either end.

What you can do once it’s connected

• Ask your AI to find notes about a topic. It searches semantically across your vault and pulls back the most relevant
• Ask it to summarize a folder, or chain related notes together
• Have it draft a new note and save it directly into your vault (with your approval)
• Use Engram’s search inside your AI workflow without exporting, copy-pasting, or duplicating your notes anywhere

You can ask Claude “what did I write about X last month?” and it actually finds the answer by calling Engram’s search tool.

The protocol, briefly

For power readers, here’s the technical surface. You can skip this and still use Engram fine.

Engram speaks MCP over HTTPS. Specifically:

• Authorization: OAuth 2.1 with Dynamic Client Registration (DCR), so clients don’t need pre-issued credentials.
• Discovery: GET /.well-known/oauth-authorization-server and tools/list so clients learn what’s reachable.
• Tools: search, get_note, write_note, list_folders, and more.
• Resources: notes addressable as URIs.
• Scopes: granular permissions you grant per session (see Scopes).

Pages in this section

• MCP Config: JSON snippet + path table for users connecting via config files instead of the GUI
• Scopes: what each permission grants
• Troubleshooting: when the connector won’t talk to Engram

For per-client setup (Claude, Cursor, ChatGPT, and more), see AI Integrations.

The interactive tool catalog (every tool Engram exposes, with input schemas and examples) ships in a later phase and will be generated directly from the backend code so it can’t drift.