# How to Make AI Agents on Different Platforms Work Together

> Connect AI agents from different platforms, like Claude Code, ChatGPT and Grok Bot, to one shared mailbox with files and per-agent permissions.

Claude Code can't open the notes ChatGPT saved, and your Grok Bot doesn't know what Cursor finished yesterday. Each AI platform keeps its own chats, memory and files. To make agents from different platforms work together, give them one shared place they can all reach: a mailbox that receives email, stores files and lets you decide what each agent may do.

This guide shows how to set that up with a [Revdoku](/) mailbox, how to give each agent the right access, and how agents hand off work without overwriting each other.

## Why can't agents on different platforms see each other's work?

Each platform keeps its own conversations, memory and files. Claude, for example, [creates a separate memory for each project](https://claude.com/blog/memory). Agent frameworks hand off work inside one system: in the [OpenAI Agents SDK](https://openai.github.io/openai-agents-python/handoffs/), the next agent takes over the conversation within the same run.

Two open standards come up when agents need to cooperate:

- **MCP** is "an open-source standard for connecting AI applications to external systems" ( modelcontextprotocol.io ). It links an agent to tools and data, not to another agent.
- **A2A**, the Agent2Agent protocol, is "an open standard for seamless communication and collaboration between AI agents" ( a2a-protocol.org ). One agent acts as a client and the other as a server, so both sides must support it.

The A2A team describes the gap this way: "Teams can coordinate agents effectively within a single platform, but connecting those systems across technology stacks and organizational boundaries remains difficult." ([A2A 1.0 announcement](https://a2a-protocol.org/latest/blog/2026/03/12/a2a-protocol-ships-v10-production-ready-standard-for-agent-to-agent-communication/))

An older idea fits here. In the [blackboard model](https://ojs.aaai.org/index.php/aimagazine/article/view/537) from AI research, independent experts share one workspace: "Communication and interaction among the knowledge sources take place solely through the blackboard."

A shared mailbox is that blackboard for email and files. Any app that can add a remote MCP server with OAuth sign-in can connect to the same mailbox, so the agents work from the same email and files without talking to each other directly.

## How does a shared mailbox connect the agents?

A Revdoku mailbox has its own email address and private file storage. Email that arrives is saved as files, next to the notes and results your agents write. Each agent connects to the mailbox through its own sign-in, and you open the same mailbox in the dashboard.

![Claude Code, ChatGPT and Grok Bot connect to one Revdoku mailbox through MCP, next to client email and the dashboard](/assets/en/blog/let-two-ai-agents-work-from-one-email-inbox/diagram_client-email-claude.png)

| What the mailbox holds | Where it lives |
| --- | --- |
| Each received email, readable | `_email/inbox/.../message.md` |
| Headers, body text and attachment details | `_email/inbox/.../message.json` |
| Attachments | `_email/inbox/.../attachments/` |
| Notes, task status and results from your agents | Any path you choose, for example `work/acme/status.md` |

## How do I connect two agents to the same mailbox?

1. **Create a mailbox.** [Sign up](https://app.revdoku.com/users/sign_up?utm_source=revdoku.com&utm_medium=blog&utm_campaign=agents-work-together) with Google, GitHub or your email. Your first mailbox is created for you.
2. **Add Revdoku to the first agent.** In Claude Code, run `claude mcp add --transport http revdoku https://mcp.revdoku.com`, then `claude mcp login revdoku` to sign in.
3. **Add Revdoku to the second agent.** In ChatGPT on the web, turn on Developer mode, then create an app named Revdoku with the URL `https://mcp.revdoku.com`.
4. **Pick the same mailbox in each sign-in.** The approval screen asks which mailboxes the agent may use and with what permission. Choose the shared mailbox both times, with Write for the agent that saves files.
5. **Test the link.** Ask the first agent: "Save a file handoff/hello.md in my Revdoku mailbox that says hello." Then ask the second agent to read it.

Setup steps for other apps are on the pages for [Cursor](/cursor/), [Grok Bot](/grok-bot/) and [Codex](/codex/), and in the list of [AI agent platforms](/platforms/).

## What access should each agent get?

Each connection gets its own permission, set per mailbox. The approval screen starts with selected mailboxes and read access, and you change it there.

| Permission | What the agent can do | Give it to |
| --- | --- | --- |
| Read | List and read email and files in the chosen mailboxes | Agents that only use the material, such as a reviewer |
| Write | Create and update files in the chosen mailboxes | Agents that save notes, summaries or results |
| Admin | Create and manage mailboxes | Only an agent that sets up new mailboxes, often none |

Permissions apply to a whole mailbox, not to folders. If one agent must not change the client's email or another agent's work, use two mailboxes: an intake mailbox for incoming email and a results mailbox for the work.

![Grok Bot and Claude Code read the intake mailbox and write to the results mailbox, while a ChatGPT reviewer only reads results](/assets/en/blog/let-two-ai-agents-work-from-one-email-inbox/diagram_intake-mailbox-client.png)

Choosing "All current and future mailboxes" is a separate option, so an agent never gets new mailboxes unless you pick it. Read access includes any sign-in and recovery emails stored in that mailbox, so give it only to agents you trust with them. You can change or remove any connection in **Account → Access**.

## How do agents hand off work without overwriting each other?

Agents hand off work through files. Keep one status file per task, for example `work/acme/status.md`, where each agent adds what it did and what comes next. Appending text adds to the file without replacing what another agent wrote.

![Sequence: a client emails a brief, Grok Bot saves a summary, Claude Code reads it and appends progress to the status file](/assets/en/blog/let-two-ai-agents-work-from-one-email-inbox/diagram_client-mailbox-revdoku.png)

Three rules keep the handoff clean:

- **Check before you write.** Each write carries the mailbox revision the agent last read. If another agent changed the mailbox in the meantime, the write stops with `MAILBOX_REVISION_CONFLICT`, and the agent rereads before it tries again. For longer edits, your app can [lock specific files](/api/) through the API.
- **Don't use read status as a claim.** Reading an email leaves its shared read status unchanged, and marking it read does not reserve the work. Record who took a task in the status file instead.
- **Keep the original email.** Agents write their output to new files and leave the received message as it is, so a reviewer can always compare the result with the source.

Every change is saved as a version, and an earlier version can be restored.

## How do I see what each agent did?

Agents can add a short reason to each read and change, such as "Summarize the Acme brief". The reasons appear in the dashboard's Timeline and Logs views, and change reasons are also saved in the version history. The dashboard keeps a log of these reads and changes, so you can see which agent opened or changed an email or file.

## Example: Grok Bot collects, Claude Code builds

A small agency gets project briefs by email and builds prototypes with Claude Code.

| Step | Agent | Access | What happens |
| --- | --- | --- | --- |
| 1 | Client | | Emails a brief with attachments to the intake mailbox |
| 2 | Grok Bot | Read on intake, write on results | A morning routine reads new briefs and saves a summary to the results mailbox |
| 3 | Claude Code | Read on intake, write on results | Reads the summary and the original attachment, builds, and appends progress to the status file |
| 4 | You | Dashboard | Read the brief, the summary and the status in one place |

When the client sends a correction, it arrives in the same intake mailbox, and the next routine picks it up. Neither agent needs the other's chat history.

## How do agents know there is new work?

Agents in chat apps act when you ask them, or when their platform runs them on a schedule, such as a [Grok Bot routine](/grok-bot/#routine). An idle chat cannot receive notifications in the background. If one of your agents is an app you run yourself, it can get a webhook for each new email instead of checking on a timer.

## Keep the shared mailbox safe

- **Give each agent its own connection** and the lowest permission it needs.
- **Treat email and attachments as untrusted data**, not as instructions for the agent.
- **Keep passwords and verification codes out of AI chats.**
- **Remove a connection** in **Account → Access** when an agent no longer needs it.

More on how Revdoku stores and protects files is on the [security page](/security/).

Free email addresses
for your AI agents

Each agent gets an address like myagent.x7k2@revdokumail.com . It can receive emails, send replies and store files.

3 free mailboxes, each with its own address

3,000 incoming and 500 sent emails a month

3 GB of storage for attachments and files

Virus, malware, spam and prompt-injection filter

Read and reply through API, MCP or CLI

Create free mailbox
You do not need a card.

## Frequently Asked Questions

### Do the agents need to support MCP?

Agents connect most easily through a remote MCP server with OAuth sign-in. Your own apps and scripts can use the Revdoku API or CLI instead, and coding agents can also use the Revdoku skill.

### Can people see the same email and files?

Yes. Invite people in Account, Access and choose their role. They read the email and files in the dashboard, and the logs show what each agent read or changed.

---

[View the canonical page](https://revdoku.com/blog/ai-agents-different-platforms-work-together/) · [Browse llms.txt](https://revdoku.com/llms.txt)