For Claude Managed Agents
A team chat for your Claude Managed Agent.
You built the agent on Claude Managed Agents. Melony lets your teammates use it: they sign in with Google, chat, and share conversations, while your Anthropic API key stays on our server.
Set up in a few minutes
Paste your Anthropic API key
Sign in with Google and choose Connect agent. Melony recognizes the Anthropic key and lists the Managed Agents and environments it can see.
Pick the agent and environment
Choose your agent and the environment its sessions should run in. A sole environment is selected for you; you can also paste the IDs.
Invite your team
Add up to 3 teammates by email. People without an account join automatically the first time they sign in with that Google email.
Chat and share
Each chat starts private. Share one with everyone on the agent when it’s useful to the team.
What stays on Anthropic, and what Melony adds
Melony is a distribution layer, not an agent builder. It never asks for your model, instructions, prompts or tools.
| Anthropic keeps | Melony adds |
|---|---|
| The agent: model, system prompt and tools | A hosted chat UI with streaming replies and chat history |
| Environments and vaults | Google sign-in and a member list per agent |
| Sessions, events and memory | Private-by-default chats you can share with the team |
| Permission policies and billing | A server-side proxy so the key never reaches a browser |
Details that matter to the agent owner.
Read the security overview for the full picture of what Melony stores.
- Environment by IDManaged Agents sessions run in an environment, so each connected agent names one. Owners and admins can switch it for new sessions; Melony never creates or edits environments.
- Vaults by ID onlyAttach an Anthropic vault by its ID and Melony passes it when creating new sessions. Secret values stay with Anthropic.
- Permissions live on the agentMelony has no approval UI. Tool calls that pause for approval are denied with a pointer to the agent’s permission policy, so a session never hangs waiting on a browser. Configure permissions in the Claude Console.
- Tool activity you can followTool use and MCP tool use appear inline next to their results. Melony displays them; the agent runs them on Anthropic.
Frequently asked questions
How do I share a Claude Managed Agent with my team?
Connect the agent to Melony with your Anthropic API key and choose its environment, then invite up to 3 teammates by email. They sign in with Google and chat with the agent in a hosted UI. Your key stays on Melony’s server.
Do my teammates need access to the Claude Console?
No. Teammates only need a Google account. Requests go through Melony’s server using the owner’s Anthropic key, which is never sent to a browser.
Why does Melony ask for an environment?
Claude Managed Agents sessions must run in an environment. Melony stores the environment ID you choose and passes it when it creates each new session. Existing sessions keep the environment they started in.
What happens when a tool needs approval?
Melony doesn’t approve tool calls. A call that pauses for approval is denied with a message pointing to the agent’s permission policy. Allow the tools your team needs in the agent’s permission policy on Anthropic.
Are there usage limits?
Each agent can have up to 3 teammates besides the owner, pending invites included, and each person can send up to 100 messages per day. Anthropic usage is billed to the owner’s key.