Akili
Docs navigation

API & MCP servers

Tools over a typed boundary: connect the systems that hold your real work, and drive the platform programmatically.

MCP in one paragraph

The Model Context Protocol (MCP) is an open standard for giving AI systems tools and data sources through a typed interface. In Akili, MCP is how agents reach beyond the chat window: query a database, read a GitHub issue, create a Jira ticket. Because the boundary is typed, an agent sees exactly the operations a server exposes — nothing more.

The apps catalog

Akili ships MCP-backed connectors you can enable without running anything yourself: GitHub, GitLab, Linear, Jira, Notion, Google Drive, Airtable, Asana, Slack, Microsoft Teams, PostgreSQL, MySQL, MongoDB, Fetch, and Puppeteer. Connector credentials are encrypted at rest, like every other secret in the platform.

Connecting your own MCP server

  1. Add the server under Settings → MCP servers (Pro and up) with its endpoint and credentials.
  2. Grant it to the agents that should be able to use it — tools are per-agent, not global.
  3. The agent's tool calls now appear in its conversation transcripts, reviewable in the agent inbox.

API keys

Pro plans and up can generate API keys to drive the platform programmatically — chat with agents from your own code, or wire Akili into systems the catalog doesn't cover.

Example — invoke an agent
curl https://api.akili.io/v1/agents/{agentId}/chat \
  -H "Authorization: Bearer $AKILI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"message": "Summarize open incidents from this week"}'

Developers building for the marketplace get a separate Publisher REST API with webhooks and developer keys — documented in the in-app developer portal under Build.

Code execution

Agents with code execution enabled run in an isolated sandbox — not in the platform's own process. Marketplace agents that request execution are additionally screened by static analysis before listing; see the marketplace review process.