Twin Me Up
Local work twin that gives every AI agent your confirmed guidance and explicit grants
- Pricing
- Open source
- Platforms
- macOS, Linux, CLI
- Runs
- Locally
- Version
- 0.2.1
- Listing updated
- today
Similar MCP servers
About Twin Me Up
Twin Me Up is an open-source MCP server for ai models and llm tooling. It runs locally on your machine and is started by your AI client.
Agents here propose, and only you decide. A twin is a local SQLite store holding your guidance, grants and candidate lessons, and the passphrase set at init time is never given to agents. Confirming lessons, granting authority, exporting and erasing all refuse to run without it. Actions in the reserved twin namespace cannot be granted to anyone.
The workflow is a loop. At the start of work the agent calls twin_brief to get your guidance for that kind of task. Before anything consequential, such as sending, paying, publishing or deleting, it calls twin_authorise, which answers allow only when a live grant covers the action and otherwise returns ask. The agent then prepares a draft and requests approval. Corrections you make are recorded as candidate lessons.
Optional pre-action hooks for Hermes, Claude Code and pi block or escalate ungranted tool calls and fail closed. The README documents thirteen agent systems, ten verified live, and says data stays on your machine with no telemetry. The author states it is not a chatbot of you or an autonomous replacement.
Install
Run in a terminal
claude mcp add twin-me-up -- uvx twin-me-upAdd to claude_desktop_config.json
{
"mcpServers": {
"twin-me-up": {
"command": "uvx",
"args": [
"twin-me-up"
]
}
}
}Add to ~/.cursor/mcp.json
{
"mcpServers": {
"twin-me-up": {
"command": "uvx",
"args": [
"twin-me-up"
]
}
}
}Add to .vscode/mcp.json
{
"servers": {
"twin-me-up": {
"type": "stdio",
"command": "uvx",
"args": [
"twin-me-up"
]
}
}
}Add to ~/.codeium/windsurf/mcp_config.json
{
"mcpServers": {
"twin-me-up": {
"command": "uvx",
"args": [
"twin-me-up"
]
}
}
}Add to your client's MCP settings (most use this shape)
{
"mcpServers": {
"twin-me-up": {
"command": "uvx",
"args": [
"twin-me-up"
]
}
}
}Using several servers? Build one config file with the MCP config generator.
Configuration
TWINMEUP_HOMEDirectory holding your twin. Defaults to ~/.twinmeup. Run `twinmeup init` there first.
Tools this server provides (5)
Taken from the project documentation; the running server may differ.
- twin_brief
- Returns your confirmed guidance and current authorisations for a workflow.
- twin_authorise
- Checks whether a live grant covers an action; answers allow or ask.
- twin_request_approval
- Asks you to approve a prepared action.
- twin_approval_status
- Reports whether an approval request has been approved.
- twin_record_correction
- Records a correction you made so it can become a candidate lesson.
What it does
- Briefing at start of work
- twin_brief returns confirmed guidance, boundaries and escalation rules for a workflow.
- Grant-based authorisation
- Actions are allowed only under explicit, scoped grants that can expire; otherwise the answer is ask.
- Approval requests
- Agents draft an action and wait for your approval before acting.
- Lessons you ratify
- Corrections become candidate lessons that only you can confirm, revise or retire.
- Enforcement hooks
- Optional hooks outside the model block ungranted tool calls.
- Local and exportable
- SQLite store with owner-only files, tamper-evident audit log, export and erasure.
Good for
- Keep one set of working preferences across several agents
- Require approval before an agent sends, pays or publishes
- Capture corrections so an agent stops repeating a mistake
- Test whether a twin beats a generic assistant on held-out decisions
Pricing
Open source
Open-source, runs locally
Twin Me Up FAQ
How do I install it?
Run uv tool install twin-me-up or pipx install twin-me-up, then twinmeup init in your own terminal.
Can an agent confirm its own lessons?
No. Only you can confirm, reject or retire lessons, and the commands need your passphrase.
Where is my data kept?
In a local SQLite store, by default in ~/.twinmeup. The README says there is no telemetry.
Which platforms are tested?
macOS and Linux are tested; Windows should work but is untested.
Is Twin Me Up free?
Yes. Twin Me Up is open source, so you can use and self-host it for free.
How do I add Twin Me Up to Claude Code?
Run this command in your terminal: claude mcp add twin-me-up -- uvx twin-me-up
What are the best alternatives to Twin Me Up?
The closest alternatives to Twin Me Up are Codebase Memory, XERJ, headroom, omniget and next-ai-draw-io.
Where this comes from
- Description: written from the project's README.
- Install steps: from the official MCP Registry entry.



