MCP server
Bring Patricia into the client you already use.
Your workspace context, files, memory, and real delegated work, inside the AI tool you spend your day in.
- 1
Copy the Patricia connector URL
One address, every client.
https://api.patricia.app/v1/mcp - 2
Open your client's connector settings
Paste the URL where your client adds connectors or MCP servers. The exact clicks for each client are one scroll down.
- 3
Sign in and start
Approve the connection in the browser window that opens. No API key, no token to paste.
No Patricia account yet? Book a demo and we set you up.
Per client
The exact clicks, for your client.
Every option lands on the same secure Patricia MCP server.
Install in Claude Code
Add Patricia from your terminal. Then sign in from Claude Code.
- Run this command in your terminal.
Terminal command
claude mcp add --transport http patricia https://api.patricia.app/v1/mcp- Open Claude Code and enter the command below.
- Select Patricia and sign in with your Patricia account.
Claude Code command
/mcpInstall in Claude.ai or Claude Desktop
Add Patricia as a custom connector from the Claude settings.
- Open Customize > Connectors.
- Select the + button.
- Select Add custom connector.
- Enter the Patricia MCP server address.
- Select Add.
- Sign in with your Patricia account.
MCP server address
https://api.patricia.app/v1/mcpInstall in Cursor
Use the install button, or add Patricia to your Cursor configuration.
Opens Cursor on this computer and asks you to confirm the server. Then sign in when Cursor asks.
Add to CursorManual setup on another device or without the button
Save this block in ~/.cursor/mcp.json, then sign in when Cursor asks.
~/.cursor/mcp.json
{
"mcpServers": {
"patricia": {
"url": "https://api.patricia.app/v1/mcp"
}
}
}Install in ChatGPT or Codex
Connect from ChatGPT settings, or add Patricia with the Codex CLI.
ChatGPT desktop app
- Open Settings.
- Select MCP servers.
- Select Add server.
- Choose Streamable HTTP.
- Enter the Patricia MCP server address.
- Save the server.
- Restart the ChatGPT desktop app.
- Sign in with your Patricia account when ChatGPT asks.
MCP server address
https://api.patricia.app/v1/mcpChatGPT on the web
- Open Settings > Security and login.
- Turn on Developer mode.
- In Plugins, select the plus button.
- Create a developer-mode app for your remote MCP server.
- Name the app Patricia.
- Enter the Patricia MCP server address.
- Choose OAuth.
- Sign in with your Patricia account when ChatGPT asks.
MCP server address
https://api.patricia.app/v1/mcpCodex CLI
- Run both commands in your terminal.
- Sign in with your Patricia account in the browser window that opens.
Add Patricia
codex mcp add patricia --url https://api.patricia.app/v1/mcpSign in
codex mcp login patriciaInstall in any MCP client
Use Patricia with any client that supports Streamable HTTP.
Choose the block that matches your client.
With OAuth
{
"mcpServers": {
"patricia": {
"type": "http",
"url": "https://api.patricia.app/v1/mcp"
}
}
}Without OAuth
{
"mcpServers": {
"patricia": {
"type": "http",
"url": "https://api.patricia.app/v1/mcp",
"headers": {
"Authorization": "Bearer pat_mcp_..."
}
}
}
}What your client can do
The context of a real teammate.
Read what Patricia knows, add to it, and hand her whole jobs. She runs them with your workspace's skills and connected tools.
Know the workspace
Confirm which workspace and which person the credential names, read the company profile, and list the team.
Read and write the workspace memory
Search what Patricia already knows about the company, store a new fact, or forget one.
Work with files
List, search, read, and create files in the workspace, and open or close a dated public link for one.
See what Patricia can already do
List the skills and the connected tools this workspace holds, then search or inspect a capability before calling it.
Hand Patricia a job
Ask Patricia to do the work as the person the token names. The run is asynchronous, so poll it for status. Patricia's ordinary approval rules still apply, so anything that publishes, sends, or spends waits for a human yes.
From Claude, for example
Ask Patricia for last month's ad numbers and a draft client update.
ask_patricia · run started
Done. Spend, results, and the draft are in your files, ready for your approval.
Bring your own servers
MCP runs both ways here.
Run a tool the integrations directory does not cover? Connect its MCP server from the console and Patricia treats it like any other tool: same permissions, same approvals, same audit trail.
Connect it your way
Sign in at the tool server's own page and approve Patricia. If a server will not register her automatically, an admin pastes a client ID and secret instead.
Team-wide or personal
An admin connects a server for the whole workspace. Any member can also connect one just for themselves: only they can use it, and it goes away if they leave.
Approvals that make sense
Plain lookups run without nagging you. Anything that changes data asks first, no matter what the server claims about itself. You can allow one specific action to run automatically, and nothing more.
Looking for the tools Patricia connects natively? Browse the integrations directory.
Access and security
Scoped, revocable, and human-approved.
pat_mcp_
Personal token
Names the person who minted it. Required for anything that reads or writes that person's own work: files, memory, skills, connected tools, and asking Patricia to do a job.
pat_live_
Workspace API key
Names the workspace and no person. It can read workspace-level context such as the company profile and the team roster. Anything that depends on who is asking is refused.
One workspace at a time
Every answer stays scoped to the one workspace behind the credential. Patricia does not read another workspace through the same connection.
A human keeps the final say
When your client hands Patricia a job, her normal approval rules apply. Work above that bar waits for a human yes in Patricia.
OAuth 2.1 with dynamic registration and S256 PKCE. Discovery: authorization server and protected resource. Revoke a token in Developer settings. See the MCP registry descriptor, the security overview, or the full developer reference.
FAQ
Got questions left?
What is MCP?
The Model Context Protocol is an open standard that lets AI tools share context and tools with each other. Patricia speaks it over Streamable HTTP, in both directions: your client can use Patricia, and Patricia can use tool servers you bring.
Which clients work with Patricia?
Claude on the web, Claude Desktop, Claude Code, Cursor, ChatGPT, Codex, and any MCP client that supports remote servers. OAuth clients connect with the URL alone; the rest use a personal token.
Do I need an API key?
No. Clients that support MCP OAuth connect by signing in, nothing to paste. A client without OAuth uses a personal token you mint once in Developer settings.
Do I need a Patricia account first?
Yes. There is no self-serve signup: book a demo and we set your workspace up together, with free credits to start.
What does it cost?
There is no separate MCP price. Work Patricia does through MCP uses your workspace credits, the same metering as chat.
Can my client change things without me?
No. Patricia's approval rules follow the work wherever it comes from: anything that publishes, sends, or spends waits for a human yes, and every answer stays scoped to the one workspace behind the credential.
Can Patricia use my own tools too?
Yes. Connect any MCP tool server to Patricia from the console, team-wide through an admin or private to just you, and she uses it under the same permissions, approvals, and audit trail.
Everybody deserves a Patricia.
She handles the work your team never gets to. You build the company. See what changes in your first week.