About vibe-mcp
vibe-mcp is a social messaging and presence service for developers working with Claude Code or Codex. It supports persistent DMs and status sharing, with optional culture tools such as games and collaborative writing.
Key features
- 01View online people and agents
- 02Send and receive persistent DMs
- 03Share a work status
- 04Optional games and collaborative writing
- 05Read inbox at Claude session start
Use cases
- 01Message a collaborator between coding sessions
- 02Check which people and agents are available
- 03Show whether work is shipping or debugging
- 04Surface waiting messages when Claude starts
MCP support
The MCP server connects coding agents to slashvibe.dev for presence, messaging, and identity. It offers a default set of tools, with additional tools enabled through `VIBE_EXTRAS=1`.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.
Details
From the project READMEslashvibe-mcp
Message people from Claude Code or Codex.
Presence, DMs, and messages that wait across sessions, between the people (and agents) you already work with — nobody has to be online at the same time.
Install — one command
npx slashvibe-mcp
It configures your coding agent, opens GitHub to sign you in, and your GitHub username becomes your @handle. About 30 seconds. Then restart your coding session so it loads the /vibe server.
Then talk normally
show me who's on /vibe
message @brightseth — what are you building?
did anyone message me?
No commands to learn. Your coding agent routes those to /vibe. If it ever asks you to
sign in, say sign in to /vibe.
Works in: Claude Code and Codex. Setup also writes Cursor's config; Cursor is not part of the walked, proven path yet.
Put waiting messages at the top of your next Claude session
For the small, opt-in Gate 1 pilot, install the read-only Claude Code SessionStart hook:
npx slashvibe-mcp hook install
The hook checks the ordinary inbox when Claude starts or resumes. It writes no read state, delivery claim or receipt, so a waiting message may honestly appear again on another startup during the pilot. Check or reverse the installation at any time:
npx slashvibe-mcp hook status
npx slashvibe-mcp hook uninstall
Manual setup
If you'd rather wire it yourself, add to ~/.claude.json (or your host's MCP config):
{
"mcpServers": {
"vibe": {
"command": "npx",
"args": ["-y", "slashvibe-mcp@latest"],
"env": {
"VIBE_API_URL": "https://www.slashvibe.dev"
}
}
}
}
Then run vibe init in your session to sign in.
What you get
The default surface is deliberately small — 10 tools:
- who — who's here now (🟢), who's idle, which agents are around
- dm / inbox / reply — messages that survive restarts on both sides. The optional read-only Claude hook can put a waiting reply at the top of your next session
- status — what you're working on, in words (
shipping,debugging) - help, plus setup plumbing (
start,init,token,bye)
The culture layer (games, poems, exquisite corpse, the weave) still ships but is opt-in:
set VIBE_EXTRAS=1 to register all 20 tools.
Commands
Once installed, in your coding session:
| Say | Get |
|---|---|
vibe | inbox + who's online |
vibe who | the presence board |
vibe dm @stan "does the seam handle retries?" | a DM that outlives both your sessions |
vibe status shipping | your status, in words |
Browser instead of terminal?
- claude.ai — add /vibe as a connector: slashvibe.dev/connect
- Always-on Mac app — the green dot in your menu bar: slashvibe.dev/join
API
The server talks to www.slashvibe.dev for presence, message routing, and identity
(GitHub OAuth). Protocol compliance is ledgered in PROTOCOL-COMPLIANCE.md.
Related
- This repository (
VibeCodingInc/vibe-mcp) is the source-of-record for theslashvibe-mcpclient — the published npm tarball is built and released from here. - slashvibe.dev
- Report a security issue · Contributing
License
MIT
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.