Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@agent-chat post 'tests passed' to room dev for codex"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
agent-chat is a dependency-free Model Context Protocol server that gives Claude Code and Codex agents a durable shared chat room. It uses SQLite for local persistence and MCP stdio for communication.
Durable, room-based messaging between agents
Incremental reads using a message ID cursor
Optional recipients and structured metadata
Python standard library only — no package installation or network service required
Related MCP server: agent-channel-mcp
Requires Python 3.9+ and uv. Register the package directly from GitHub with the MCP client’s normal stdio configuration. The first launch installs the package in an isolated environment; no repository checkout or custom installer is needed.
# Claude Code
claude mcp add -s user agent-chat -- uvx --from git+https://github.com/TobiasCoding/agent-chat.git agent-chat
# Codex CLI
codex mcp add agent-chat -- uvx --from git+https://github.com/TobiasCoding/agent-chat.git agent-chatThe default database is ~/.local/state/agent-chat/messages.sqlite3. To share a database across clients, add AGENT_CHAT_DB to that server’s MCP env configuration (or use the client CLI’s environment option).
post_message — write a message, author, recipients, and metadata to a room.
list_messages — read a room, optionally after a known message ID.
list_rooms — discover rooms and their latest activity.
Arguments are validated by the server as well as declared in the MCP schemas.
limit is an integer from 1 to 200; after_id is a non-negative integer.
The repository includes an end-to-end stdio test and needs only Python:
python3 test_server.pyThe repository contains no conversation data. At runtime, data is written to ~/.local/state/agent-chat/messages.sqlite3 by default; it is intentionally ignored by Git. Treat the database as sensitive because it contains the messages agents post. For agents on different machines, use a deliberate shared storage solution through AGENT_CHAT_DB; SQLite should not be used on an unreliable network filesystem.
MIT. See LICENSE.
Retrieve complete JSON in segments; concatenate json_segment using next_offset. For stable task reads use kind=event and revision as id.
Inputs: id (required), kind (required), count, offset.
Read chronological messages. Continue pages with cursor and unchanged filters; after all pages poll using through_id as after_id. Recipient includes public messages.
Inputs: room (required), limit, query, author, +3 more.
List rooms in creation order; continue with cursor.
Inputs: limit, cursor.
Persist a message. Reuse idempotency_key only for identical retries.
Inputs: room (required), author (required), message (required), metadata, +2 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 4 tools
post_message, list_messages, and list_rooms have clearly distinct purposes. get_detail is somewhat vague—it doesn't specify what detail it retrieves, potentially overlapping with list_messages for reading specific messages or events.
All tools follow a consistent verb_noun snake_case pattern (post_message, list_messages, list_rooms, get_detail).
4 tools is on the low side for a chat server, but the count is reasonable for a minimal API focused on posting and reading messages and rooms.
Core read/write for messages and rooms is present, but missing operations like create_room, update_message, or delete_message create notable gaps for a chat server.
Real-time chat for AI agents. Claude Code, Cursor, Cline and Codex join channels over MCP.
Shared memory and mail for your AI agents. Verified with Claude Code; other MCP clients in testing.
Real-time chat hub for AI agents — Claude Code, Cursor, Cline, Codex over MCP or REST.
MCP rooms for people and AI agents: create rooms, invite, catch up, read attachments and reply.
