Persome

What Persome MCP server does

The Persome MCP server creates a local memory layer for AI agents running on a Mac. It builds an evolving Personal Model from focused activity captured after the user grants the required macOS permissions. The resulting context is intended to help trusted agents understand current goals, priorities, working patterns, decision preferences, and unfinished work.

The model is organized into progressively broader structures: Points represent sourced observations or events; Lines connect relationships or changes over time; Faces describe patterns supported by related evidence; Volumes group higher-order structures; and the Root represents the current integrated model. Claims retain source receipts so they can be inspected, corrected, revised, exported, or deleted.

This places Persome in the 🧠 Knowledge & Memory category. It is designed for developers who want the same personal context available across multiple MCP-compatible coding agents rather than maintaining separate agent histories.

How it works

The runtime operates locally on macOS and is private by default. During onboarding, it explains permission requests before they appear. Accessibility is granted to the versioned mac-ax-helper and, when event-driven capture is enabled, mac-ax-watcher. Screen Recording is requested only when the effective screenshot or local-OCR policy requires pixel access.

After capture and processing, connected agents can query the model through MCP. Agents select and execute tasks; the runtime supplies evidence-linked context rather than acting as an autonomous task executor.

Setup and configuration

Persome requires macOS 13 or newer and the Xcode Command Line Tools. The package-managed installation uses the personal-model distribution and installs the persome command:

uv tool install personal-model
persome onboard
persome model open --after 30

A source-based alternative is to clone the repository and run bash install.sh. Successful interactive onboarding opens the unified local setup experience. The persome onboard step is also where the permission boundary is explained and established.

Connection commands are available for supported clients. Examples include persome install claude-code, persome install codex, persome install cursor-agent, and persome install claude-desktop. For other compatible clients, the installer can generate an MCP JSON file with persome install mcp-json --filename persome-mcp.json.

Tools and capabilities

The Persome MCP server supports these documented capabilities:

• Provide evidence-linked local context through MCP.
• Let agents search the Personal Model for unfinished work and relevant context.
• Help agents compare proposed next steps with recorded priorities.
• Preserve one model for Claude Code, Codex, Cursor Agent, Claude Desktop, and other compatible clients.
• Let the owner inspect, correct, export, or delete the model and its data.

The project describes other MCP servers, including Filesystem, GitHub, Slack, and Google Drive, as separate tools rather than built-in Persome integrations.

Limitations and notes

The Persome MCP server is intended for macOS and requires macOS 13 or newer. It depends on explicit operating-system permissions, and some capture features require Accessibility, Screen Recording, or both according to the selected policy.

It is a server for trusted MCP clients, not a replacement for those clients' task execution. External actions remain subject to the user's authority. The project is licensed under Apache-2.0.