Connect Your Tools to idapt Over MCP

MCP (Model Context Protocol) is the open standard that lets AI tools plug into data sources and capabilities once, instead of per-app integrations. idapt speaks it as a server: any MCP client can reach your workspace.

What connecting gets you

Your MCP-capable tools (Claude Code, editors, custom agents) gain idapt's capabilities as native tools:

Drive: search, read, and write your files from the client you are in.
Agents and memory: run your agents, query the shared memory they keep.
Chats: search history, start conversations, continue threads.
Computers: run commands on machines you have paired, permissions willing.

The concrete magic is shared state: an agent in your editor and a chat in idapt reading the same memory and the same files stop being two AI silos.

Setup

Create an API key scoped to what the client should reach (Settings → API Keys).
Add idapt's MCP endpoint to your client's configuration; the MCP page has the exact snippet per client.
The client discovers the available tools automatically; call them like any built-in.

Permissions are the key's, not the protocol's: a read-only key gives every client read-only reach, whatever it asks for.

Good workflows

Editor-driven research: from Claude Code, search your idapt Drive for the design doc and pull the relevant section into the session.
Cross-tool memory: your terminal agent records decisions to an idapt agent's memory; tomorrow's chat session already knows them.
One knowledge spine: keep sources and outputs in Drive; let every MCP client read the same spine instead of amassing per-tool copies.

Good to know

What is MCP explains the protocol itself.
The help articles cover setup per client, available capabilities, and permission mapping.
The same surface exists as a CLI and SDK: one contract, three transports.

If you use Claude Code, do the five-minute setup and try "search my idapt drive for X" from a coding session: cross-tool workspaces stop being theoretical the first time it answers.