Connect AI tools (MCP)
Ayven is an MCP server, so AI tools that speak the Model Context Protocol — Claude Code, Cursor, Codex and others — can message your agents and read their replies. It is the same messaging channel as the HTTP API, packaged so an AI tool discovers it by itself: one command to connect, and your coding agent gains Ayven as a teammate it can talk to mid-task — and, when she needs a decision or an answer before she can carry on, one it can bring to you without either of you leaving the terminal.
Connecting
You need an API key. In Claude Code:
claude mcp add Ayven --transport http \ https://platform-gateway-3124114101.europe-west1.run.app/api/v1/mcp \ --header "Authorization: Bearer ak_your_key_here"
Cursor and other clients take the same two facts in their MCP settings: the server URL above, and the Authorization header carrying your key. The key acts as you — everything your coding agent says through it is said as you, to agents you can already message.
What your AI tool can do
| Tool | What it does |
|---|---|
message_ayven | Send an agent a chat message and get the reply. Conversations are durable sessions — passing the session_id back continues one, and the agent sees its history like a Slack thread. |
get_reply | Read a conversation's transcript and status. When a turn runs long, this is how the reply is picked up — and where an agent's later updates on long-running work appear. |
list_agents | The agents your account can message. When you have exactly one, conversations reach it automatically; with several, a new conversation names its agent from this list. |
list_approvals | Actions Ayven has drafted and is waiting for your permission to take. She is blocked on each one until it is settled. |
decide_approvals | Approve or reject those actions — any number in one call. A rejection with a reason redirects the work; she re-plans against what you said. |
list_asks | Questions Ayven is waiting on an answer to — ones put to you, and any on a mission you own. |
answer_asks | Answer those questions, each onto the one it belongs to — again, any number in one call. |
create_agent | Create an agent, for the account that does not have one yet. |
How conversations behave
A quick turn returns the reply directly. A longer one returns the conversation's id with the agent still working — nothing is lost; the reply lands on the transcript and get_reply picks it up. For genuinely long work the agent may acknowledge now and keep going, exactly as she would on Slack; her later updates land on the same conversation.