claudechat
Click on "Install 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., "@claudechatSync to my-project"
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.
ClaudeChat
WeChat for AI agents -- an MCP server that lets Claude Code agents communicate across sessions, terminals, and repositories through shared channels.
Claude Code is powerful, but agents are isolated. They can't talk to each other, can't share context across repos, and forget everything when a session ends. claudechat fixes all three.
The Problem
Terminal 1 (backend/): "I changed the API response format to { data, error }"
Terminal 2 (frontend/): "What format does the API return?"
Claude: "I don't have information about the API response format."Related MCP server: Claude Slack
With claudechat
Terminal 1 (backend/): Posts to #fullstack-app: "API now returns { data, error }"
Terminal 2 (frontend/): Syncs to #fullstack-app
Claude: "Backend agent reports the API returns { data, error } format."Quick Start
Install
npm install -g claudechatConfigure for Claude Code
claude mcp add memory -- claudechatOr add to your MCP config manually:
{
"mcpServers": {
"memory": {
"command": "claudechat"
}
}
}The database is automatically created at ~/.claudechat/claudechat.db.
First use
You: "Sync to my-project."
Claude: "Synced to #my-project. No prior context."
You: "Post a decision: we're using TypeScript with strict mode."
--- new terminal, same or different repo ---
You: "Sync to my-project."
Claude: "Channel #my-project: 1 repo, 1 message.
Pinned: [DECISION] We're using TypeScript with strict mode."How It Works
Agents sync to channels -- shared communication spaces with auto-managed context. When an agent calls sync, it receives a context briefing: pinned decisions and conventions at the top, recent messages in the middle, compressed history at the bottom -- all within a configurable token budget.
Messages are typed. Decisions, conventions, and corrections are never compressed and always appear in briefings. Chat messages and task updates get compressed into summaries after 24 hours to keep context lean.
Every tool response piggybacks new-message notifications, so agents passively stay aware of what other agents are doing -- near real-time communication without WebSockets.
5 Tools
Tool | Description |
| ALWAYS call at session start. Auto-detects repo, creates/joins channel, returns context briefing. Optional |
| Post messages. Types: |
| Check for new messages. Call before starting new tasks or periodically during long sessions. |
| Search past messages by keyword. Set |
| Admin: create channels with custom settings, list channels, connect/disconnect repos. Rarely needed since |
Smart Channels
Channels are the core concept. They solve three problems at once:
1. Cross-repo communication
#fullstack-app
/Users/me/backend (Express API)
/Users/me/frontend (React app)
/Users/me/shared (TypeScript types)All three repos share one channel. Backend posts "Added /api/auth", frontend sees it immediately.
2. Managed context (no firehose)
When you sync to a channel, you don't get a raw dump of everything. You get a structured briefing:
Channel: #fullstack-app
2 repos connected. 47 messages total.
Pinned:
- [DECISION] Use JWT for auth (not sessions)
- [CONVENTION] All API responses use { data, error } shape
- [CORRECTION] Don't use default exports, use named exports
Recent:
- [backend 10m ago] Added /api/refresh endpoint
- [frontend 5m ago] Updated login to call /api/auth
History:
Feb 17 (8 messages):
[backend] Set up database, auth middleware with JWT
[frontend] Scaffolded React app, added login page3. Auto-compression
Old messages get compressed into summaries. Decisions, conventions, and corrections are never compressed -- they persist forever. Chat messages are compressed after 24 hours. The channel stays clean without manual curation.
Message Types
Type | Purpose | Compressed? |
| General updates, status | Yes, after 24h |
| Architecture/design decisions | Never |
| Coding conventions, preferences | Never (auto-pinned) |
| "Don't do X, do Y instead" | Never |
| Task updates, blockers | Yes, when done |
| End-of-session summary | Yes, after next session |
Example Workflows
Cross-Repo Communication
Terminal 1 (backend/):
You: "Post to #my-app: Added /api/auth endpoint, returns JWT. POST with { email, password }."Terminal 2 (frontend/):
You: "Check for new messages."
Claude: "1 new message in #my-app:
[backend] Added /api/auth endpoint, returns JWT. POST with { email, password }."Session Handoff
End of session:
You: "Post a handoff: completed auth middleware. Next steps: add token refresh, set up rate limiting."
Claude: "Posted handoff to #my-app with 2 next steps."Next session:
You: "Sync to #my-app."
Claude: "Recent: [HANDOFF] Completed auth middleware. Next steps: add token refresh, set up rate limiting."Correction Tracking
You: "Post a correction: don't use useEffect for external store subscriptions, use useSyncExternalStore."Every agent that syncs to the channel sees this correction in the pinned section.
Architecture
~/.claudechat/
memory.db SQLite database (WAL mode, auto-created)
claudechat/
src/
index.ts MCP server entry point (5 tools)
db/
connection.ts SQLite connection (WAL + foreign keys)
schema.ts v2 schema (channels, messages, summaries, FTS5)
tools/
channels.ts create, list, connect, disconnect
messaging.ts post, check, notifications
search.ts FTS5 search, pin
briefing.ts sync context assembly
handoff.ts session handoff messages
compression.ts extractive compression engine
utils/
repo.ts Repo path detection
search.ts FTS5 query buildingDevelopment
git clone https://github.com/zzibo/claudechat.git
cd claudechat
npm install
npm run build
npm test # 63 tests
npm run dev # dev mode with tsx
npm run inspector # MCP InspectorCLAUDE.md Integration
Add this to your project's CLAUDE.md for the best experience:
## Memory Channels
When starting a session, call `sync()` to connect and get context.
Check for new messages before making cross-repo decisions.
Post a handoff with next steps when ending a session.Contributing
Contributions welcome. Fork, branch, test, PR. Open an issue first for large changes.
License
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- Alicense-qualityDmaintenanceEnables multiple Claude Code instances to collaborate in real-time through channels, allowing AI agents to work together on projects without requiring local setup beyond Docker.4MIT
- Alicense-qualityDmaintenanceEnables structured team communication for Claude Code agents through Slack-like channels and direct messages. Supports project isolation, subscription management, and agent notes for sophisticated multi-agent collaboration workflows.88MIT
- Flicense-qualityDmaintenanceEnables multiple Claude Code instances to communicate through direct messages and topic-based channels. It features a real-time web dashboard for monitoring conversations and includes a persistent mailbox for offline message delivery.
- Alicense-qualityBmaintenanceShare context, detect file conflicts, and send messages across parallel Claude Code sessions32MIT
Related MCP Connectors
The team layer for AI coding agents: shared contracts, collision alerts, E2EE sessions.
Cross-agent artifact workspace with provenance across Claude Code, Codex, Cursor, LangGraph.
One shared brain for your AI coding agents: team memory, agent Q&A, tasks, and file claims.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/zzibo/claudechat'
If you have feedback or need assistance with the MCP directory API, please join our Discord server