EmperorClaw MCP
MCP server of the self-hosted EmperorClaw panel: Claude Code and Codex read company knowledge, tasks and customers in scope
- 698 ★ on the repository
- Last commit
Broad access
Why this level
- Panel agents run in containers that share the host kernel, and over MCP they can write tasks, files and notes
- The panel stores company knowledge, customers and spend, and an MCP token opens everything in its scope
- Agents run around the clock and spend model provider money until the budget limit triggers
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/emperorclawDetects 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.
- Panel agents run in containers that share the host kernel, and over MCP they can write tasks, files and notes
- The panel stores company knowledge, customers and spend, and an MCP token opens everything in its scope
- Agents run around the clock and spend model provider money until the budget limit triggers
This is third-party code. Review the repository files before installing.
What it does
EmperorClaw is a control panel for a team of AI agents that you run on your own server through Docker Compose with PostgreSQL. It has an agent org chart, a task board, approvals, budgets with an automatic pause and spend tracking per run, agent and customer. For outside assistants the panel answers MCP at `/mcp`: Claude Code, Codex and Claude Desktop connect through OAuth 2.1 with PKCE or with a token from settings. Over MCP an assistant reads company knowledge, tasks, projects, customers, agents and files, but only those visible to its connection. Scope is set per customer or agent, an empty scope grants nothing, and writes follow the same rules. The README describes 38 MCP tools, and workflow templates come from `/api/mcp/templates`. Agents inside the panel run on Hermes or OpenClaw, each in its own Docker container. You can ask Claude Code for the delivery rules of one customer and the open tasks on their project without exposing the data of other customers.
Who it is for. Agencies and small teams where several AI agents work for different customers and each customer's data has to stay separated.
Good fit when
- Claude Code has to know one customer's rules and open tasks without seeing the others
- One installation serves several customers and a contractor should see only their own agents and files
- Agents run around the clock and need a monthly budget that stops them at the limit
- An agent's decisions need manual approval: payments and outgoing mail wait for a person
- Company knowledge lives in scoped notes instead of being pasted into every chat
Not a fit when
- You run a single local agent without a server, where a panel with Docker and PostgreSQL is overkill
- You need built-in agent evaluations or a prompt studio, which the README says the panel lacks
- You require an MIT license, while the project uses Fair Source FSL-1.1-Apache-2.0
Example request
Connect to EmperorClaw and tell me the delivery rules for Maple and what is still open on their projectMCP
- Transport
- http
- Authentication
- OAuth
Environment variables
EMPEROR_CLAW_TOKENsecret- Access token from Settings, Access Tokens. Needed when connecting without OAuth.
Limitations
The panel needs your own server with Docker and PostgreSQL, and the first install pulls images and takes noticeably longer than `npx`. Agent containers share the host kernel and protect files, keys and the process, but do not isolate hostile code. Each agent gets 1 GiB of memory and 1.5 CPUs. The Fair Source license (FSL-1.1-Apache-2.0) turns into Apache 2.0 after two years. There are no built-in agent evaluations, no portable company package and no native mobile app, only a responsive website.
How to disable. Disconnect MCP in Claude Code with `claude mcp remove emperorclaw` and in Codex with `codex mcp remove emperorclaw`. Revoke the token in the panel settings. `docker compose down` in `~/emperorclaw` stops the panel itself.
FAQ
How do I connect EmperorClaw to Claude Code?
Create a token in Settings, Access Tokens and run `claude mcp add --transport http emperorclaw https://your-host/mcp --header "Authorization: Bearer $EMPEROR_CLAW_TOKEN"`. Codex uses `codex mcp add emperorclaw --url ... --bearer-token-env-var EMPEROR_CLAW_TOKEN`.
Can I connect without a token?
Yes. Add `https://your-host/mcp` as a custom connector in Claude Desktop or claude.ai and approve the OAuth 2.1 sign-in.
What does an assistant connected over MCP see?
Only the knowledge, tasks, projects, customers, agents and files in the scope granted to the token or account. An empty scope grants nothing, and a restricted member cannot reach another customer by link.
How do I install EmperorClaw itself?
With `curl -fsSL https://raw.githubusercontent.com/emperorclaw/emperorclaw/main/install.sh | bash` or `docker compose up -d` from the repository. The installer writes to `~/emperorclaw`, checks ports and opens the browser.
Which agents run inside the panel?
Agents run on Hermes or OpenClaw, one Docker container per agent. MCP is the panel's extension surface.
Discussion
No questions or reviews yet.
Related
MCP serversA CLI that connects local agents to a shared cloud store of sessions, memory, skills and MCP connectors.
A read-only MCP server for the HeyClaude registry: an agent finds skills, MCP servers and hooks and gets ready install commands
PromptOps keeps Claude Code, Codex and Gemini CLI side by side by project and gives a lead agent MCP to launch helpers
A CLI that turns any MCP server, OpenAPI spec or GraphQL endpoint into commands at runtime, with no codegen
MCP server for a shared agent room: decisions, tasks verified by another agent and a final report, joined by code.
A Claude Code configuration with safeguards: approvals for risky actions, command audit, sub-agents and 65 on-demand MCP servers.
README in short
The README presents EmperorClaw as an office for AI agents: an org chart, a task board, shared company memory and spend control on your server. It details member access scopes, a knowledge base scoped by company, customer, project and agent, the MCP server and a capability table with its gaps stated. It then covers team and role templates, the Docker stack layout, a comparison with a local CLI and one-command installation.