MCP and integrations
The Model Context Protocol is an open standard for connecting tools to AI. In AION it works both ways.
A practice workspace with step-by-step guidance and result checks. No sign-up or account changes.
How it works
Agents connect to remote MCP servers over HTTP or SSE. The verified catalogue includes DeepWiki, Context7, Microsoft Learn, GitHub, Hugging Face and more.
Add your own server by URL. If it needs a token, the token is stored encrypted on the server. A connection is personal or for a project. Local stdio servers are not supported.
AION itself is an MCP server too. Connect Claude, ChatGPT or another client with an access key or by signing in with OAuth, and work with AION tasks and documents from there.
GitHub connects with a personal token: link a repository to a project and see branches and commits for tasks.
Name: DeepWiki
URL: https://mcp.deepwiki.com/mcp
Transport: httpWhy it helps
- One standard instead of a dozen separate integrations.
- Your usual AI client works with AION tasks directly.
- Tokens are never stored in the browser.
Get started
- 1
Open Integrations → MCP store and pick a server.
- 2
Or add your own by URL on the “My servers” tab.
- 3
Enable the server for an agent under “What it may do”.
- 4
To connect AION to an outside client, create a key under “Access keys” or sign in with OAuth.
Where to find it
Integrations → MCP storeIntegrations → My serversIntegrations → Access keysSpace → Settings → Integrations
Details
AION as an MCP server
- Endpoint /mcp, Streamable HTTP and SSE transport.
- Sign in with a personal key with scopes, or with OAuth 2.1 and PKCE.
- Files up to 15 MB.
Telegram
- An agent replies in Telegram once the bot is linked on the agent's card.
Write to support and we will reply by email. Or just try it: your personal space is ready right after sign-up.
Checked against the code on 4 November 2026