Connect your AI assistant
Work with outwrite.ai from an assistant you already use.
Use outwrite.ai from your favorite AI assistant or coding agent. The MCP connection lets your assistant read your workspace and perform the actions you authorize. You don't need to host a server or share your website credentials.
Get your connection URL
Open AI connections in outwrite.ai and copy Your MCP server URL. Use the URL shown in the workspace you want to connect. Production and staging are separate.
Connect Claude
- Open Claude's connector settings and add a custom remote connector.
- Name it outwrite.ai and paste your MCP server URL.
- Follow the authorization prompt. Sign into outwrite.ai, choose your workspace and review the permissions.
- Return to Claude and enable the connector in the conversation where you want to use it.
For the terminal agent, follow Claude Code setup below.
Connect Codex
In a terminal, replace the placeholder with the exact URL copied from your workspace:
codex mcp add outwrite --url "YOUR_WORKSPACE_MCP_SERVER_URL"
codex mcp login outwrite
Complete the browser consent flow. Start a new Codex session if the new tools aren't loaded yet. The desktop app's MCP settings can also manage configured servers.
Connect ChatGPT
In a ChatGPT workspace with custom MCP creation enabled, open Plugins, select Add custom MCP server, name it outwrite.ai and paste your workspace MCP URL. Choose OAuth authentication, review the requested access and complete sign-in. In the older Apps interface, use Apps → Create instead.
Enable the connection in a conversation and ask it to list your outwrite.ai workspaces. Workspace administrators control custom app access. ChatGPT's setup guide explains its account requirements and current settings.
Connect Claude Code
Run this command with your copied workspace URL:
claude mcp add --transport http outwrite "YOUR_WORKSPACE_MCP_SERVER_URL"
Open Claude Code, run /mcp, select outwrite and complete authentication. Then ask Claude to list your authorized workspaces.
Connect Cursor
Open Cursor's MCP settings, add a server named outwrite, and use this configuration with your copied URL:
{
"mcpServers": {
"outwrite": { "url": "YOUR_WORKSPACE_MCP_SERVER_URL" }
}
}
Complete the sign-in prompt and enable the server's tools for your agent. Cursor's MCP guide covers user-level and project-level configuration.
Connect GitHub Copilot
In VS Code, run MCP: Add Server from the command palette. Choose HTTP, paste your MCP URL and name the server outwrite. Start the server and complete browser authorization. Open Copilot's agent chat and enable the outwrite tools.
Use VS Code's MCP configuration guide for workspace configuration and tool controls.
Connect more coding agents
These clients use the same workspace URL and browser sign-in. Add a remote Streamable HTTP server, rather than a local server command. After connecting, ask the agent to list your authorized workspaces before making changes.
| Client | Where to connect |
|---|---|
| Windsurf / Cascade | Open Cascade's MCP configuration and add outwrite with your workspace URL. Follow the Cascade MCP guide for its current configuration format. |
| Cline | Open MCP Servers, add a remote server and enter your URL. See Cline's MCP guide. |
| Gemini CLI | Add an HTTP MCP server with your URL in the CLI's MCP configuration, then authenticate through /mcp. See the Gemini CLI guide. |
| Kiro | Add outwrite to the MCP configuration with a url entry. Kiro opens the browser authorization flow. See Kiro's configuration guide. |
| OpenCode | Add outwrite as a remote MCP server with your URL and enable it. Complete OAuth when prompted. See OpenCode's MCP guide. |
| Goose | Add a remote extension using Streamable HTTP and your URL, then complete sign-in. See Goose's extension guide. |
| Roo Code | Open MCP Servers, add a Streamable HTTP server with your URL and authorize the connection. See Roo Code's MCP guide. |
Connect your agent
For OpenAI Dots, Grok Bot or an agent with its own tool setup, give it the following connection brief along with your copied MCP URL:
Connect outwrite.ai as a remote Streamable HTTP MCP server using this workspace URL. Use OAuth browser sign-in and let me choose the workspace and permissions. After authorization, list my workspaces and summarize my plan. Do not change anything yet.
Use the agent's remote MCP connection settings or its supported MCP bridge. This is a connection to the agent's tools, not a setting on the underlying AI model. You complete the authorization yourself; never paste passwords or access tokens into the conversation. If your client needs a local bridge, follow its maintained setup guide rather than disabling authentication.
Choose what your assistant can do
Start with read access for reporting. Add management permissions for scheduling, preparing or editing articles, Brand changes and supported work controls. Publication needs a separate permission and still follows article checks and your saved publishing policy.
A connection is limited to your authorized workspace and permissions. It does not expose other customers or stored integration credentials. You can revoke it in AI connections and reconnect with different permissions. Your assistant manages its own conversation history.
Permissions and troubleshooting explains every scope, previews, revisions and error responses. You don't need to learn the tool reference for everyday use.
Try it out
Begin with: “List my authorized outwrite.ai workspaces, then summarize the selected workspace's plan.” This confirms that the connection can read your account, rather than merely appearing in settings.
| What you want | What to ask |
|---|---|
| A briefing | “What's planned next, what needs attention, and what changed in our visibility?” |
| Answer evidence | “Show the saved answer and citations for this prompt, with its observation date.” |
| A schedule change | “Move this draft to next Tuesday at 9 AM in our workspace timezone. Show me first.” |
| A first draft | “Schedule this idea as draft-only, then prepare it.” |
| A focused edit | “Read this draft and propose a clearer opening. Keep the table and sources.” |
Scheduling creates a plan item; preparation writes it. A preview isn't an applied change, and a queued operation isn't finished work. Ask your assistant to check the final status. Explicitly authorize publication only after it has inspected the current article and destination.
The built-in Ask chat has its own supported actions. Granting an external assistant publication permission doesn't grant that permission to Ask.
Connection troubleshooting
If consent expires, start a fresh connection flow. If a tool is missing, check the permissions you granted. Confirm your client supports remote HTTP MCP with OAuth and uses the correct workspace URL.
An expired subscription or trial must be resolved in outwrite.ai. MCP does not bypass account access. Don't share callback codes or access tokens in chats or support messages.
