# phaedo-mcp [Health: Active]

**Category:** 💻 Developer Tools  
**Repository:** https://github.com/phaedo-labs/phaedo-mcp  
**GitHub Stars:** 0  
**npm Downloads (last month):** 2407  
**Views:** 0  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/phaedo-mcp-2

## Description
Carry your Phaedo cognitive fingerprint into any MCP client: injection, consults, escalation.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "phaedo-mcp": {
    "command": "npx",
    "args": ["-y","phaedo-mcp"]
  }
}
```

## Documentation

## What phaedo-mcp does

phaedo-mcp connects MCP-compatible AI clients with a portable Phaedo cognitive fingerprint. The fingerprint represents communication style, working style, and decision patterns, and is used to provide context at the beginning of a session. The server also handles consultations when an agent reaches a decision that may depend on how the user would act.

A consultation produces a signal such as `proceed`, `clarify`, `escalate`, or `decline`, along with confidence. When the available information does not support a reliable answer, the server can return an abstention signal instead of guessing. User-authored deference policies take priority over inferred behavior, and explicit standing authorizations are auditable.

## How it works

The reference server is published as `phaedo-mcp` on npm and communicates with MCP clients over stdio. Running `npx phaedo-mcp` is the general-purpose launch path. Claude Code has a plugin that includes the server and hooks, while Claude Desktop can use the packaged desktop bundle.

The full fingerprint is supplied by the Phaedo phone app and learned on-device from the user's activity. The Claude Code plugin can create a starter profile locally when the phone-based profile is unavailable. A sample profile can be selected with `PHAEDO_FINGERPRINT=sample` for testing. If no profile is available, the server reports that condition rather than generating an inferred identity.

For blocking signals, the server can contact a paired phone in real time. The user can approve, deny, or modify the requested action. Those overrides contribute to the delegated “act-as-me” behavior. Communications between a computer and paired phone are end-to-end encrypted, and the relay carrying them cannot read them.

## Setup and configuration

For Claude Code, add the Phaedo marketplace and install the Phaedo plugin:

```bash
claude plugin marketplace add phaedo-labs/phaedo-mcp
claude plugin install phaedo@phaedo
```

For Claude Desktop, download `desktop/phaedo.mcpb` and open it. Other MCP clients can launch the server with:

```bash
npx phaedo-mcp
```

Set `PHAEDO_FINGERPRINT=sample` when a made-up example profile is needed. Pairing, policy configuration, receipts, and the agreement metric are covered by the server's companion guides.

## Tools and capabilities

- Injects a cognitive fingerprint at session start.
- Answers agent consultations with calibrated decision signals.
- Applies user-defined deference policies and explicit authorizations.
- Abstains when the fingerprint lacks sufficient coverage.
- Escalates blocking decisions to a paired phone.
- Records an encrypted audit trail on the device.
- Uses user overrides to inform delegated-work behavior.

## Limitations and notes

The learning pipeline that builds fingerprints is separate from this open repository. The phone app provides the full learned profile and is not publicly available. The open portion contains the fingerprint standard, schemas, MCP binding, conformance suite, and reference server for consuming and interoperating with fingerprints.

The server exposes consultation signals rather than the internal fingerprint structure. Fingerprints are stored on user devices in encrypted form, and the server does not keep plaintext fingerprint data at rest. The Phaedo name and logo are trademarks.

_Full upstream README: https://allmcps.com/mcp/phaedo-mcp-2/readme_