What MateMCP MCP server does
MateMCP MCP server gives compatible AI clients a controlled route to work on an enrolled Windows or macOS computer. The local Agent can operate on configured project files, run shell commands, interact with browsers and supported desktop interfaces, handle approved attachments, and use managed secrets without placing those credentials in model context. Project-specific knowledge can be stored in versioned SKILL.md files, while global cross-project knowledge is retained in the Agent.
The system is intended to continue useful work when a provider’s built-in coding or agent feature reaches its usage limit. It does not raise or bypass that provider quota; it gives the conversation another MCP-based path to the user’s own machine.
How it works
An AI client connects to an Agent-specific MCP URL through the MateMCP Relay. The Relay routes traffic to the selected online Agent, while the Control Plane manages accounts, device ownership, OAuth, authorization, and remote approval coordination. The Agent performs operations locally, and Companion provides local management for status, approvals, shells, secrets, activity, diagnostics, updates, and lifecycle controls.
OAuth tokens are associated with the user, Agent or resource, and granted scopes. Capabilities such as mcp:read, mcp:write, and mcp:shell are enforced. Filesystem operations remain within configured project roots, and sensitive actions may require approval. Agent credentials and user-managed secrets are stored in macOS Keychain or Windows Credential Manager rather than exposed through the MCP URL.
Setup and configuration
Create or access an account through the MateMCP account portal, install MateMCP Desktop, and complete device enrollment in Companion when requested. Then copy the Agent’s unique MCP URL into a compatible AI client and complete OAuth with the account that owns the Agent. A safe first check is to ask the client to list a configured project or inspect a file.
On Apple Silicon macOS, the documented bootstrap script installs or upgrades Agent and Companion. On Windows x64, the corresponding PowerShell bootstrap performs the same role. The desktop installation preserves user-scoped configuration and secure credentials. Computer Use on macOS requires Accessibility and Screen Recording permissions.
Tools and capabilities
MateMCP MCP server supports the following documented capabilities:
- Read and modify files inside configured project roots.
- Run commands through shell and interactive terminal sessions.
- Keep long-lived shells and resume output after short reconnects.
- Browse websites and interact with browser or native user interfaces where supported.
- Inspect screenshots, capture screenshots, and support visual QA workflows.
- Transfer conversation attachments with bounded, resumable, integrity-checked operations.
- Inject approved secrets into shell work without revealing them to the AI.
- Request approvals and review activity, audit records, credential use, and diagnostics.
- Connect multiple independently revocable Agents under one account.
Limitations and notes
The provider’s usage quota remains unchanged. Access is limited by configured project roots, granted scopes, local permissions, and approval decisions. The Relay does not need the user’s operating-system secrets. Short Agent or Relay interruptions can be handled by logical sessions and stable operation identities, but clients may retain an older MCP tool snapshot after Agent updates; the README specifically notes a refresh procedure for ChatGPT.
macOS development or ad-hoc builds may require Accessibility and Screen Recording permissions to be granted again after some updates because production signing and TCC identity hardening are still being improved.