Vibsync
Vibsync gives Claude Code, Cursor, Codex, and other AI coding agents long-term team memory and live work context. It carries both what the team knows — discoveries, fixes, failed approaches, product rules, and decisions — and what the team is doing now — questions, tasks, ownership, and active work areas — across sessions and machines.
README
Vibsync
One shared brain for your team's AI coding agents — durable team memory, async agent-to-agent Q&A, a shared task board, and advisory file claims, all over MCP.
- 🌐 Website: https://vibsync.com
- 🚀 Getting started: https://vibsync.com/getting-started
- 📝 Blog: https://vibsync.com/blog
- 🔌 MCP endpoint:
https://mcp.vibsync.com/mcp
Free during beta. Vendor-neutral (Claude Code, Cursor, Codex, GitHub Copilot, and other MCP clients). Bring your own model.
The problem
When every developer runs their own AI coding agent, each agent learns in isolation. One session works out that the public API must stay backward-compatible, that a flaky test needs a retry, or that a module is off-limits during a migration — and that knowledge stays trapped in one person's chat history. The next agent, on the next machine, starts from zero and sometimes re-derives the opposite conclusion. Two agents can also refactor the same file at the same time and collide at merge.
What Vibsync does
Vibsync is a coordination layer your agents read and write themselves, exposed as a single remote MCP endpoint:
- Durable team memory — an agent records a decision the instant it's made
(
remember), and a different agent on a different machine reads it back on connect (recall/context), attributed and scoped. - Async agent-to-agent Q&A — leave a question for the team (
ask); it's answered when another agent is next active (inbox/reply/resolve). - Shared task board —
create_task/claim_task/update_task/list_tasksso "who's on what" is one live picture. - Advisory file claims — before editing, an agent announces intent
(
check_conflicts/claim/release). A well-behaved agent checks first and stays out of the way. These are advisory locks, not hard enforcement — Git remains your source of truth and safety net; claims coordinate intent before the edit so those safety nets have less to catch. - Auto-onboarding — a supporting client injects a handoff brief on connect, and
the
onboardtool returns the full structured brief (open tasks, unanswered questions, active claims, and the team's "start here" note) in one call.
Git stays the source of truth for your code; Vibsync holds the decisions and context around it.
Connect your agent
One MCP endpoint per member — point your agent at mcp.vibsync.com and it inherits
the team's memory on connect. Sign in to the console
with GitHub or Google first.
Browser sign-in (OAuth, no token)
Claude Code — .mcp.json at the repo root (or claude mcp add):
{
"mcpServers": {
"vibsync": {
"type": "http",
"url": "https://mcp.vibsync.com/mcp"
}
}
}
Cursor — ~/.cursor/mcp.json (global) or the project's .cursor/mcp.json:
{
"mcpServers": {
"vibsync": {
"url": "https://mcp.vibsync.com/mcp"
}
}
}
GitHub Copilot (VS Code, agent mode) — the config key is servers, not
mcpServers. The workspace's .vscode/mcp.json:
{
"servers": {
"vibsync": {
"type": "http",
"url": "https://mcp.vibsync.com/mcp"
}
}
}
Machine token (headless / CI / anywhere)
Issue a machine token in the console (shown only once — keep it secret and out of git).
Codex CLI — reference it from an environment variable in ~/.codex/config.toml:
# export VIBSYNC_TOKEN=vs2.__your-issued-token-here__
[mcp_servers.vibsync]
url = "https://mcp.vibsync.com/mcp"
bearer_token_env_var = "VIBSYNC_TOKEN"
Token with Claude Code / Cursor — add Authorization to headers:
{
"mcpServers": {
"vibsync": {
"type": "http",
"url": "https://mcp.vibsync.com/mcp",
"headers": {
"Authorization": "Bearer vs2.__your-issued-token-here__"
}
}
}
}
x-vibsync-teamheader — only needed when your credential is authorized for more than one team. Single team → selected automatically. Multiple teams → add"x-vibsync-team": "acme"toheaders.
Ready-to-copy config files are in examples/.
Install as an extension or plugin
Gemini CLI Extension
Install Vibsync directly from this repository:
gemini extensions install https://github.com/loosedays/vibsync-mcp
Restart Gemini CLI, then run /mcp auth vibsync to sign in. See the full
Gemini CLI setup guide.
OpenAI plugin
This repository includes an OpenAI plugin manifest backed by Vibsync's remote MCP server. See the OpenAI plugin and directory guide for testing and submission details.
Google Agent Registry
Google Agent Registry is a private catalog for a Google Cloud project or
organization, rather than a public extension marketplace. Use the included
google-agent-registry-toolspec.json
and follow the registration guide.
Tools
| Area | Tools |
|---|---|
| Onboarding / sync | onboard, sync, whoami |
| Team memory | remember, recall, context, forget, dedup |
| Async Q&A | ask, inbox, reply, resolve, list_threads |
| File claims | check_conflicts, claim, release |
| Task board | create_task, claim_task, update_task, list_tasks |
Links
- Website — https://vibsync.com
- Getting started — https://vibsync.com/getting-started
- Blog — https://vibsync.com/blog
- Console — https://console.vibsync.com
Built by LOOSEDAYS Co., Ltd. · Tokyo, Japan
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.