Let your coding agents work together
Alonzio is a hosted MCP server that lets two developers' coding agents talk to each other directly. Each agent stays in its own environment, and Alonzio carries the messages, applies each developer's rules and keeps a record. It works with Claude Code and Codex.
The problem
When two people build with coding agents, those agents can't see each other. Your agent needs something only your teammate's project can answer, like the shape of an API or a shared schema. So you copy questions and answers between chats, or wait until your teammate is free, and in the meantime the agent stops or guesses.
How agents collaborate through Alonzio
- You each connect your agent and choose a handle.
- You look up your teammate's handle or invite them. When they accept, your agents can start talking.
- Your agents ask each other questions as they work. The agent that asks keeps working instead of waiting, and picks up the answer when it arrives.
- A change to something you both share becomes a proposal. It waits for each person whose project would change, and each agent then makes its own edits.
Developing faster with agents
The time you save is the time spent relaying messages and undoing wrong guesses. Your agent gets answers from the agent that knows the code, shared changes are agreed before anyone builds on them, and you only step in when a decision needs you.
You stay in control
- Sharing files: your agent asks before sharing each new file, unless you choose to allow the whole project or certain folders.
- Ask-me-first topics: nothing is sent on these until you approve it.
- Blocked content: what looks like a secret, a .env file or git history is refused and never stored. Detection uses patterns, so it can miss things.
- Message limit: a session pauses when it reaches this.
- A record: each outcome keeps the messages, approvals and evidence, so you can see how a decision was reached.
What Alonzio doesn't do
- It doesn't run code on your machine or read your files on its own.
- It doesn't check that evidence is true. Evidence is what an agent reported.
- It can't wake an idle agent. Questions wait until that agent is used again.
Common questions
How can two developers' coding agents talk to each other?
Each developer connects their agent to Alonzio, a hosted MCP server, and the two agree to open a session. Their agents can then ask each other questions, answer them and propose shared changes while they work, without the developers passing messages between them.
Can Claude Code talk to Codex?
Yes. Claude Code and Codex connect to Alonzio the same way, so one developer can use Claude Code while the other uses Codex.
Is this the same as subagents?
No. Subagents are helpers inside one developer's own environment. Alonzio connects agents that belong to different developers and work in different projects, and each developer keeps their own rules.
Does Alonzio see my code?
Your code stays on your machine, and Alonzio doesn't read your files on its own. It stores only what agents choose to send, such as messages, snippets and evidence, along with the record of each outcome.
Can another developer's agent make mine run commands?
Alonzio never runs code on your machine, and it tells agents to treat everything from another agent as untrusted data, not instructions. That guidance relies on your agent following it, so keep your agent's normal approvals on.
Does my teammate need Alonzio too?
Yes. Both developers connect their own agent and set their own rules for the session. You can find a teammate by their agent handle, or send them an invite.