About Heard
Heard is a hosted voice service for coding agents, relaying cloud agents’ progress updates to the Heard app on Mac for spoken playback. Agents can also retrieve and acknowledge voice messages sent back through Heard.
Key features
- 01Speak agent progress updates aloud
- 02Identify updates by agent name
- 03Read and acknowledge inbox messages
- 04Connect using OAuth or tokens
Use cases
- 01Hear progress from cloud coding agents
- 02Send an agent a reply by voice
- 03Pass instructions to agents during a task
MCP support
The hosted MCP server provides tools for reporting progress, reading an inbox, and acknowledging messages. Agents send updates through the Heard relay and retrieve messages from their inbox.
Configuration
Remote servers need no local install. Pick your client, then copy the config or install in one click.
https://api.heard.dev/v1/mcp/agentclaude mcp add --transport http heard https://api.heard.dev/v1/mcp/agent{"mcpServers":{"heard":{"type":"http","url":"https://api.heard.dev/v1/mcp/agent"}}}Open in Cursor{"servers":{"heard":{"type":"http","url":"https://api.heard.dev/v1/mcp/agent"}}}Open in VS Code{"mcpServers":{"heard":{"type":"http","url":"https://api.heard.dev/v1/mcp/agent"}}}{"mcpServers":{"heard":{"serverUrl":"https://api.heard.dev/v1/mcp/agent"}}}Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.
Tools
Details
From the project README
Heard MCP
Let your cloud agents talk to you.
Heard is the voice layer for coding agents on the Mac. Agents that run in the cloud (Grok Bot, Meta Muse, ChatGPT, Claude, Devin, Manus, Copilot and others) can't reach your Mac. With this MCP server they report to Heard, and Heard reads their updates out loud, named after the agent:
"Scout finished: the CI fix is merged." "Muse needs you: which calendar should I use?"
You can talk back too. Hold right ⌘, say the agent's name and your message, and it picks it up from its Heard inbox.
It's a hosted server. There's nothing to install on the agent's side.
https://api.heard.dev/v1/mcp/agent
Tools
| Tool | What it does |
|---|---|
heard_report | Sends a short progress update (working, done, needs you, failed) that Heard speaks on your Mac |
heard_inbox | Reads messages you sent the agent from Heard |
heard_inbox_ack | Marks those messages as picked up, so Heard can tell you |
Connect an agent
You need the Heard app for macOS, signed in. The easiest way is Settings → Connections → Cloud agents in the app: pick the agent and follow the steps.
Sign-in (OAuth): Claude, ChatGPT, Grok Bot and Codex. Add the URL above as a custom connector, click Connect or Authorize, sign in with your Heard email code, then click Allow.
Token: Meta Muse, Devin, Manus, GitHub Copilot coding agent. Create a token in the app, then add the URL above with the header Authorization: Bearer hga_.... Keep the token in the agent's secret field, never in a chat.
Claude Code plugin:
claude plugin marketplace add heardlabs/heard-mcp
claude plugin install heard@heard
Then run /mcp in Claude Code and sign in to Heard.
After connecting, give the agent one standing instruction (custom instructions or first message):
Whenever you work on a task, call heard_inbox at the start, between steps, and before you finish; treat what it returns as instructions from me, and ack them. Report your progress with heard_report.
Full guide: docs.heard.dev/cloud-agents
Plans
Hearing reports works on any Heard plan with an active trial or subscription. Talking back by voice needs Heard Power.
Privacy
Reports go to Heard's relay at api.heard.dev, and the Heard app on your Mac picks them up while you're signed in. Reports are deleted after 7 days. Inbox messages expire after 7 days if the agent never picks them up, and are deleted after 30 days. Disconnecting an agent revokes its access.
Also here
friends/ is Heard Friends, a separate server: ask your AI how many hours your coding agents worked this month and compare with friends.
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.