Lobu MCP
Shared company context for agents over MCP: an event graph with permissions, approvals and audit
- 234 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The agent gets shared company context from Slack, GitHub, CRM and other connected sources, which can hold customer data
- Lobu specialists and tools perform actions in connected systems, so permissions and approvals are set at connection time
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/lobuDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent gets shared company context from Slack, GitHub, CRM and other connected sources, which can hold customer data
- Lobu specialists and tools perform actions in connected systems, so permissions and approvals are set at connection time
This is third-party code. Review the repository files before installing.
What it does
Lobu is an open event-sourced context layer for AI agents. Every tool call, data change, device signal, API event and conversation becomes an event linked to the people, projects, customers and companies it belongs to. The agent gets a live queryable model of the world instead of a transcript it forgets after a session. ChatGPT, Claude, Codex and your own agents read and write the same history with permissions respected. The MCP server is available at `https://lobu.ai/mcp`: any client connects without the Lobu runtime and without a `lobu.config.ts` file. The commands `npx @lobu/cli@latest connect claude-code`, `connect codex` and `connect opencode` detect the client, install MCP and a skill bundle or suggest the steps. Sign-in goes through OAuth, and the sources you need are connected and opened for sharing. There are also persistent specialists for a durable responsibility: support, releases, incident follow-up, talked to in the web app or Slack and called by other agents over MCP.
Who it is for. Teams where several agents and people work with the same customers, tasks and data and each agent needs a shared context with checkable permissions.
Good fit when
- Claude Code and ChatGPT should see one picture of a customer instead of rebuilding context from Slack, GitHub and CRM again
- A support agent pulls the earlier history of requests and decisions before replying
- A long responsibility such as release coordination is handed to a persistent specialist that other agents call
- You need approvals and a log to understand where the agent took data and who allowed an action
Not a fit when
- A single agent in a single session: a shared context layer gives little and complicates the setup
- You cannot connect internal systems to an external service: the hosted `lobu.ai` address takes data out of your environment
- You need simple note-style memory: lightweight memory servers are enough for that
Example request
Connect to Lobu and show what is known about the customer Acme: recent requests, related tasks and who handled them, then suggest the next stepMCP
- Transport
- http
- Authentication
- OAuth
Limitations
Agent access is set by the permissions granted when sources are connected. For a local run `lobu run` starts a stack with embedded Postgres and a web interface on port 8787, and an external Postgres needs `DATABASE_URL` with pgvector. The `init` command creates `ENCRYPTION_KEY` in `.env`, and with manual setup you supply the key yourself: 32 bytes in base64, and replacing the key makes earlier data unreadable. The MCP registry server version is 0.2.0 and the canary CLI downloads components on first run. The license is Apache-2.0.
How to disable. Disconnect the server in the client and revoke access in the Lobu settings. The local stack is stopped by ending `lobu run` and data lives in your database.
FAQ
How do I connect Lobu to Claude Code?
Run `npx @lobu/cli@latest connect claude-code`, complete OAuth sign-in and connect the sources you need. For Codex and OpenCode use `connect codex` and `connect opencode`.
Is the Lobu agent runtime required?
No, any MCP client connects to `https://lobu.ai/mcp` without the Lobu runtime or a `lobu.config.ts` file.
How do I run it locally?
With `npx @lobu/cli@latest init my-specialist`, then `run`: the stack starts with embedded Postgres and an interface on port 8787. For an external database set `DATABASE_URL` with pgvector.
What is a persistent specialist?
An agent for a durable responsibility: customer research, support triage, release coordination or incident follow-up. People talk to it in the web app or Slack and other agents call it over MCP.
Which clients are supported?
Claude Code, Codex, OpenCode, Antigravity, ChatGPT, Claude Desktop, Cursor and custom MCP clients. Guides for Claude, ChatGPT, Codex and Grok are on lobu.ai.
Discussion
No questions or reviews yet.
Related
MCP serversMCP server of the self-hosted EmperorClaw panel: Claude Code and Codex read company knowledge, tasks and customers in scope
A universal MCP CLI client from Apify: persistent sessions, OAuth, JSON output for scripts and shell access to any MCP server for agents
A CLI that turns any MCP server, OpenAPI spec or GraphQL endpoint into commands at runtime, with no codegen
An MCP server that lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
An MCP server that runs sub-agents defined in Markdown through Codex, Claude Code, Cursor CLI and other CLIs from any MCP client
The mcpl CLI searches tools across several MCP servers and calls them from the terminal, with OAuth sign-in done by a human
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Lobu (this card) | yes | OAuth | 234 | |
| MCP Memory Service | no | not required | 2,117 | standalone agent memory without a shared model of people, customers and systems with permissions and approvals |
| agentmemory | yes | not required | 30,314 | memory for a single agent rather than a shared company event graph |
README in short
The text explains the idea of a shared context layer: each agent gets a live model of your world instead of re-fetching from Slack, GitHub and CRM. Next come a ChatGPT demo, a quick start for Claude Code, Codex and OpenCode, persistent specialists, the CLI with a TypeScript SDK and the platform design.