Connect your agents to the projects your team keeps in AUVY.
Through the MCP server at api.auvy.ai, Claude, ChatGPT, Cursor and Codex search a project and read and save its text files. Your own code calls the API on the same host and runs AUVY's agent, which names the file and version behind every answer. Workspace data is stored on our servers in Helsinki.
The docs at docs.auvy.ai open once you sign in with an AUVY account.
What you can build
What an agent can do in a project.
- Work in the files
- Your agent searches a project, reads its text files such as Markdown, CSV and JSON at any version, and saves new ones. Other files open in AUVY, and your own code fetches them through the API. AUVY keeps every earlier version and shows who or what changed it.
- Run AUVY's agent
- Start a run in a project from your code and follow each step as it streams. Every answer names the file and version it came from, and when a step needs a person, the run stops and asks them by name.
- Put a bot on a project
- A bot joins a project like a member, with its own identity, memory and connected accounts, and works under that project's permissions.
- Use Microsoft 365
- Signed in as you, your agent reads the Outlook mail and calendar you connected to AUVY. Its search also finds the SharePoint and OneDrive files placed in a project.
Connect
One address for every agent.
https://api.auvy.ai/mcp- Claude, ChatGPT and Claude Code
- Add the address above as a connector and sign in with your AUVY account. The agent then acts as you and opens only the projects you can open. In Claude Code, one command adds it:
claude mcp add --transport http auvy https://api.auvy.ai/mcp - Cursor, Codex and VS Code
- Add the address above with a workspace key as the bearer token. The key is bound to the project that is open when a workspace admin makes it, and acts there for the workspace. Admins make keys in AUVY under Settings, Keys.
- Your own code
- Call api.auvy.ai with a workspace key. Every error has the same JSON shape, and runs stream as server-sent events.
Where data goes
Where your project data is processed.
Workspace and account data sit on our servers in Helsinki, and your data never trains a model. Mail sent into a project arrives through Resend, a US company, and some service metadata is processed outside the EU.
An agent you connect, such as Claude or ChatGPT, reads only what its sign-in or key may read, and what its vendor then does with it falls under your own terms with that vendor. Acting as you, an agent can also send mail and change your calendar; AUVY marks those tools so your client can ask you first. Every subprocessor and region ↗
Status
What is open today, and what comes next.
- API
- Agents and runs, bots, files, meetings and connected accounts at api.auvy.ai, called with a workspace key. Anyone can read the contract: OpenAPI 3.1
- Open
- MCP server
- One address for Claude, ChatGPT, Claude Code, Cursor, Codex and VS Code. Its tools and how an agent signs in: developers.md, auth.md
- Open
- Console
- Keys, usage per key and the setup for each agent, in one place inside AUVY for workspace owners and admins. Until it opens, a workspace admin makes keys under Settings, Keys.
- Not open yet
- Docs
- Guides and the full reference at docs.auvy.ai, open to anyone signed in with an AUVY account.
- After sign-in
- AUVY plugin
- One package for Codex, Claude Code and Cursor that will show your agent where each project stands, with sources, and what changed against what was agreed. It is to be listed in the Codex plugin directory first, and until then you connect the MCP server.
- Planned
- TypeScript SDK
- A typed client for the API, to be published on npm as auvy.
- Planned
