Connect Patcherly to your AI assistant¶

Patcherly offers a MCP (Model Context Protocol) server so AI tools such as Cursor, VS Code, Claude Desktop, and other MCP-compatible clients can work with your workspace — list errors, review sites, export metrics, and (when permitted) approve fixes — using the same permissions as the dashboard.

Before you connect¶

1. A workspace admin must enable MCP for your workspace under Profile → MCP (Read / Write / Export toggles per domain). New workspaces start with MCP off until an admin opts in.
2. You need a Patcherly account and access to the workspace you want the assistant to use.
3. Your plan limits (fixes, sites, exports, etc.) still apply — MCP uses the same quotas as the dashboard.

[!TIP] Switch to the correct workspace in the dashboard before you authorize MCP. The token is bound to the workspace that was active at consent time. Changing workspaces later does not widen an existing token — disconnect and authorize again if you need a different workspace.

What you can do with MCP¶

When your role and workspace policy allow it, your assistant can:

Never available via MCP: account deletion, billing or payment changes, reading raw AI API keys, MFA enrollment, inviting team members, or wiping entire databases. Destructive deletes require an explicit confirmation flag in the tool call — your assistant should ask you before proceeding.

Connect your client¶

Production MCP URL:

https://mcp.patcherly.com/mcp

Authorization (browser sign-in) uses your Patcherly dashboard at https://app.patcherly.com (authorize page). When your client opens OAuth, approve only the scopes your workflow needs. After you approve, the browser may hand off to the assistant and leave the tab open — once the client shows connected, you can close that window.

Quick install¶

Step-by-step install for each app: MCP client setup (troubleshooting stays on this page).

Workspace admin — MCP policy¶

Workspace admins configure MCP under Profile → MCP:

• Read — list and inspect data in that domain.
• Write — approve fixes, change error status, manage sites, and similar dashboard actions exposed as MCP tools.
• Export — CSV exports via export_targets, export_errors, export_metrics, export_usage, and export_audit (same data as dashboard Export buttons).

If all domains are off, authorize requests return forbidden until an admin enables at least one domain.

Tightening policy later revokes scopes that exceed the new policy on live tokens automatically.

Manage connections¶

• Profile → MCP (AI assistant connections) — see your active MCP sessions (one row per assistant connection; optional assistant label, scopes, last used) and revoke any you no longer trust.
• Password reset revokes MCP tokens for your account (same hygiene as other OAuth sessions).
• Removed team members lose MCP access immediately; re-invited users must connect again.

Example prompts¶

Once connected, you can ask your assistant in natural language, for example:

• “List open errors on my WordPress site.”
• “Summarize fix usage this billing period.”
• “Show audit events from the last 24 hours.”
• “Approve the pending fix for error ID …” (only if Write is enabled and you confirm)

The assistant discovers available tools from the MCP server after authentication. If a tool is missing, your role or workspace policy may not include that domain.

Security tips¶

• Grant only the scopes you need on the consent screen.
• Revoke unused tokens from Profile → MCP.
• Do not paste Patcherly passwords or connector secrets into AI chat — MCP uses OAuth, not your connector HMAC secret.
• Treat write-capable tokens like dashboard admin access — revoke them when a project ends.

Troubleshooting¶

For connector setup (errors from your app), see Installing a Connector.