Connect your AI

Avera is an MCP server. Any MCP-capable client can connect to a project and work the record with the same tools Avera’s built-in chat uses: Claude Code, Claude Desktop, ChatGPT, or your own agents.

Connector URL

https://app.avera.io/mcp

The endpoint is global: which tenant and which projects a connection may touch is chosen during sign-in and consent, never in the URL. Your client discovers the OAuth configuration automatically.

The operating model

• Reads are scoped. A connection is authorized for one tenant and a chosen set of projects, and never exceeds the live access of the person who approved it. Revoking that person’s access ends the connection’s reach on the next request.
• Every write is a proposal. A connected agent cannot change the record directly: creates, edits and links land as pending proposals, and a named human reviews and accepts or dismisses each one in Avera. The review link for each proposed batch is returned to the client.
• Everything is attributable. A connection acts as “client, acting as person” in the history, and every accepted change carries that provenance.

Connecting

• Claude Code: claude mcp add --transport http avera https://app.avera.io/mcp
• Claude Desktop: Settings → Connectors → Add custom connector, with the connector URL above.
• ChatGPT: Settings → Connectors → Add a custom connector (developer mode), with the connector URL above.
• Your own agent: any MCP client library; the endpoint speaks streamable HTTP with OAuth 2.1.

Whichever client: it starts an OAuth flow, you sign in, pick the tenant, select the projects and scopes (read, propose, generate) the connection may use, and approve. The client receives a scoped token; tokens rotate automatically and are revocable in Avera at any time.

No API key is involved, and the connection needs no AI key in Avera: your client brings its own model.

The tools

Every tool’s reference is generated from the running backend and cannot drift from the deployed surface. The tools group into four pages:

• Project & context: choosing the project and branch, loading the project’s summary, conventions and settings, and the caller’s review queue.
• Content: the meta-model brief, search and reads over items, documents and relationships, traversal, and authoring as proposals.
• Configuration: layouts, and configuration changes proposed into change requests.
• AI actions: the project’s configured AI actions, runnable on your client’s own model.