Connect your AI agent

Connect your AI client to Chatropic to work with your workspace through a conversation. Your agent can find knowledge articles, investigate support conversations, review content gaps, prepare documentation, and inspect billing usage. With your permission, it can also create, update, and publish knowledge articles. Chatropic MCP works with any client that supports remote Streamable HTTP MCP and the server’s OAuth sign-in flow. It is not specific to one AI provider. A client’s support for local MCP servers alone is not enough; it must support remote servers and browser-based authorization.

This guide connects your AI client to Chatropic. To give a Chatropic agent tools from another provider, see Tools and MCP.

Before you connect

You need a Chatropic account with membership in the workspace you want to use, plus an AI client that supports the connection requirements above. The permissions you can approve depend on your workspace role. You do not need to publish an agent, create a sandbox key, or provide database credentials. Widget publishable keys and other Chatropic API keys are not substitutes for the MCP OAuth connection.

Connect in your AI client

Choose your client below. Each uses the same Chatropic server and workspace consent flow.

Claude
Codex
Cursor
Other clients

For Claude on the web or Claude Desktop:

Open Customize → Connectors.
Select + → Add custom connector. Name it Chatropic and enter:
https://app.chatropic.com/mcp/chatropic
Leave optional OAuth client credentials empty, add the connector, and choose Connect.
In Chatropic, sign in, choose your workspace, and approve the permissions you need.
Return to Claude. Enable Chatropic for the conversation from + → Connectors, then try the example request below.

For a Team or Enterprise organization, an owner first adds the connector under Organization settings → Connectors → Add → Custom → Web. Each member then connects using their own Chatropic account.See Claude’s custom connector guide for current client controls and plan availability.

With Codex CLI installed, add the server from your terminal:

codex mcp add chatropic --url https://app.chatropic.com/mcp/chatropic

Complete any sign-in prompt. If authentication did not start, run:

codex mcp login chatropic

In the browser, sign in to Chatropic, choose your workspace, and approve access. Return to Codex and use /mcp to check the connection.Alternatively, add this entry to ~/.codex/config.toml, then run the login command above:

[mcp_servers.chatropic]
url = "https://app.chatropic.com/mcp/chatropic"

Codex clients sharing this configuration can use the same server entry. In the desktop or IDE MCP settings, you can also add a Streamable HTTP server named chatropic with this URL, then authenticate when prompted.See the official Codex MCP guide for client-specific settings.

Add Chatropic to ~/.cursor/mcp.json for use across projects, or .cursor/mcp.json for one project. Merge this entry into the existing mcpServers object if the file already contains servers:

{
  "mcpServers": {
    "chatropic": {
      "url": "https://app.chatropic.com/mcp/chatropic"
    }
  }
}

Open Cursor’s MCP settings, enable Chatropic, and start its OAuth connection. When Chatropic’s consent page opens, sign in, select your workspace, and approve access. Return to Agent chat and check that Chatropic’s tools appear.

Desktop sign-in limitation: Cursor desktop authorization currently requires a Chatropic OAuth compatibility update. If you receive invalid_redirect_uri, contact Chatropic support. Cursor web/Agents authorization has not yet been verified end to end.

See Cursor’s MCP guide for configuration and authentication details. A widget key or manually added Bearer header will not resolve a callback rejection.

Add a remote MCP server named Chatropic with this URL:

https://app.chatropic.com/mcp/chatropic

Choose Streamable HTTP and automatic/OAuth authentication. Start sign-in, select your Chatropic workspace, approve the requested permissions, and enable the connection for your chat or agent.Your client must support remote MCP with public-client OAuth and PKCE. Chatropic handles client registration; you do not need to create a client secret or paste a widget key.

Verify your connection

If you approved knowledge reading, try:

Find our knowledge articles about getting started. Summarize the coverage and include the article IDs. Do not change anything.

Your client should call Chatropic’s knowledge tools and return information from the workspace you selected. If the tools do not appear, refresh the client’s tool list and check that the connection is enabled for the current chat or agent.

Choose permissions for your task

Permissions apply to the selected workspace. They do not provide access to other workspaces or bypass your existing role. Chatropic checks access again when tools run, so removing someone’s workspace access also affects their MCP access.

Permission	What the client can do	Workspace access required
Agent configuration (`agents:read`)	List agents and read a limited configuration summary	View settings
Support conversations (`conversations:read`)	Search web conversations and read redacted customer-facing messages	View inbox
Support analytics (`analytics:read`)	Read aggregate support counts	View analytics
Knowledge (`knowledge:read`)	Read articles, private drafts, and detected content gaps	View sources
Draft editing (`knowledge:write`)	Read the documentation guide, inspect drafts, create articles, and update drafts	Edit sources
Article publishing (`knowledge:publish`)	Publish a specified article revision	Edit sources
Billing (`billing:read`)	Read billing summaries, usage, and invoice summaries	View billing

Workspace owners and administrators can approve all these categories. Other members see the categories allowed by their assigned permissions. A client may request fewer categories than your role permits. For a documentation assistant, select knowledge reading and draft editing. Add article publishing if you also want it to publish reviewed work. For a support review, select conversations and analytics; add knowledge reading to compare answers with your articles.

What the AI client receives

Knowledge tools return full article titles and content, including private drafts and content-gap summaries, without redaction. Granting knowledge access therefore allows the client to read unpublished material in the selected workspace. Conversation text and some agent fields are redacted. If the redaction service is unavailable, those fields can be withheld. Billing tools exclude payment details, and analytics can suppress small counts. A tool result is data for the AI client to work with. The client decides how to summarize it and may have its own confirmation settings for tool calls. In particular, granting publishing access allows publishing calls; it does not guarantee a separate Chatropic confirmation screen for each article.

Manage or disconnect access

Open Chatropic connections, sign in, and select your workspace. Review the client connections and revoke access for any you no longer need. You can also remove the connection from your AI client. To change the selected workspace or approved permissions, reconnect and complete authorization again. Revoke the previous grant if you no longer need it. Disconnecting prevents future access; it does not erase information the client already received.

Troubleshooting

Problem	What to check
The client cannot add the server	Confirm it supports remote Streamable HTTP MCP and OAuth. Use the full `/mcp/chatropic` URL, not the Chatropic home page or connections page.
Sign-in never opens	Choose automatic/OAuth authentication. If the client only accepts a fixed Bearer token, use a client with compatible OAuth support.
Your workspace is missing	Sign in with the account that belongs to that workspace. Ask its administrator to confirm your membership and MCP access.
A tool is missing	Check the approved permission categories, your workspace role, and the client’s own tool enablement settings. Reconnect to request additional permissions.
Access is denied after previously working	Your grant may have expired or been revoked, or your workspace permissions may have changed. Sign in again and review access.
You can read articles but cannot edit or publish	Knowledge reading does not include draft editing or publishing. Authorize the additional categories required for your task.
Text says “Content withheld”	Some agent and conversation fields require redaction. Knowledge article titles and content should return verbatim in the current server. Retry a fresh call; if articles are still withheld, contact support with the tool name and request ID.
An edit reports a revision conflict	Another edit changed the article. Read its current draft again, compare changes, and retry against the new revision. Do not blindly overwrite it.
A call is rate-limited or returns too much data	Pause and retry fewer requests, use smaller pages, and narrow the search. See Tool reference.

When contacting support, include the AI client name, tool name, approximate time, error, and request ID if available. Do not include passwords or access tokens. Continue with the available tools or try the example workflows.

Quick Start

Agent Management

Chatropic MCP

Actions

Settings & Guardrails

Deploy

Analytics & Monitoring

Connect your AI agent

Before you connect

Connect in your AI client

Verify your connection

Choose permissions for your task

What the AI client receives

Manage or disconnect access

Troubleshooting

​Before you connect

​Connect in your AI client

​Verify your connection

​Choose permissions for your task

​What the AI client receives

​Manage or disconnect access

​Troubleshooting

Before you connect

Connect in your AI client

Verify your connection

Choose permissions for your task

What the AI client receives

Manage or disconnect access

Troubleshooting