Connie MCP Server — Setup Guide

You will need:

• A Connie account (sign up or sign in) and membership of at least one workspace.
• An AI assistant that supports remote MCP servers over HTTPS. Every client below does.

The server address is the same for every client:

https://mcp.connie.ai

There is nothing to install: the server speaks the MCP Streamable HTTP transport directly. Authentication is OAuth 2.0 with automatic client registration. The first time a client connects, your browser opens a Connie sign-in and consent screen; approve it and the client is connected. Each person authorizes their own client, and an authorization only ever reaches the workspaces that person is a member of.

Some clients, such as Codex and Cursor, can also connect with a personal API key instead of the browser sign-in. Create one in Connie's API key settings and send it as Authorization: Bearer <key>. A key acts with your own permissions and is tied to one workspace. The Connect my agent page in Connie (in the left sidebar) shows the exact setup for each client.

Note: Connie is a remote server. Paste the HTTPS address into your client's remote, HTTP or "Streamable HTTP" option. Only a client that cannot connect to a remote server with OAuth itself (Windsurf and Cline today) needs the mcp-remote bridge, as shown below.

Adding Connie as a custom connector in Claude makes it available in Claude on the web, in the Claude desktop app and on mobile, because connectors are stored with your Claude account rather than on one machine.

Claude Pro and Max

1. In Claude, open Customize → Connectors.
2. Click +, then Add custom connector.
3. Paste the server address above as the remote MCP server URL. Leave Advanced settings empty — Connie registers the client automatically.
4. Click Add, then Connect. Sign in to Connie, pick the workspace you want as the default, and click Allow.

Claude Team and Enterprise

Custom connectors are added once by an organization owner and then connected by each member.

1. An owner opens Organization settings → Connectors, clicks Add, hovers Custom and selects Web.
2. The owner pastes the server address, clicks Add, and the connector appears for the whole organization.
3. Each member then opens Customize → Connectors, finds Connie and clicks Connect to run their own sign-in.

Using it in a conversation

In a chat, click + at the lower left, open Connectors and make sure Connie is switched on. Then ask in plain language, for example "List my Connie workspaces" or "Find every company in the Warm Leads list without a website".

The Claude desktop app uses the same Connectors settings. Connie does not go into claude_desktop_config.json — that file is for local stdio servers only.

ChatGPT connects to remote MCP servers through developer mode. It is available on Plus, Pro, Business, Enterprise and Education plans on the web.

1. Open Settings → Security and login and turn on Developer mode. On some builds the switch lives under Settings → Connectors → Advanced instead.
2. Go to the connectors (apps) section of Settings and choose Create (the + button).
3. Enter a name such as Connie, a short description, and the server address above as the MCP server URL.
4. Choose OAuth as the authentication method and create the connector. It appears under your drafts or connectors list.
5. Click Connect (or Authenticate), sign in to Connie, pick a default workspace and click Allow.

In a chat, open the + menu, choose Developer mode and select Connie for that conversation. ChatGPT asks you to confirm before it runs any tool that writes data; review the arguments it shows before approving.

Grok connects to remote MCP servers as custom connectors on grok.com.

1. On grok.com, open Connectors and add a new custom connector.
2. Paste the server address above and save.
3. Connect, sign in to Connie, pick a default workspace and click Allow.

These clients keep the connection on the machine where you add it. Each one runs the same browser sign-in the first time it needs Connie.

Claude Code

Add the server from a terminal. The default scope is the current project; add it for every project with the user scope.

# this project only
claude mcp add --transport http connie https://mcp.connie.ai

# every project on this machine
claude mcp add --transport http --scope user connie https://mcp.connie.ai

Then start Claude Code and run /mcp. Connie shows as Needs authentication; select it and follow the browser sign-in. You can also authenticate from the shell with claude mcp login connie. To share the entry with a team through version control, add it to .mcp.json in the project root:

{
  "mcpServers": {
    "connie": {
      "type": "http",
      "url": "https://mcp.connie.ai"
    }
  }
}

Useful commands: claude mcp list, claude mcp get connie, claude mcp logout connie to clear the sign-in, and claude mcp remove connie to remove the server together with its stored tokens.

Codex CLI

codex mcp add connie --url https://mcp.connie.ai
codex mcp login connie --scopes mcp:full

The login step opens the browser sign-in and stores the resulting tokens locally. The same entry can be written by hand to ~/.codex/config.toml:

[mcp_servers.connie]
url = "https://mcp.connie.ai"

codex mcp list shows configured servers and their authentication status.

Codex app and the Codex IDE extension

1. Open Settings → MCP servers (in the IDE extension, the gear menu → MCP servers).
2. Click Add server and choose Streamable HTTP.
3. Enter the name Connie and the server address, then save.
4. Choose Restart (or Restart extension). Connie shows an authentication status; select Authenticate and complete the browser sign-in.

The Codex CLI, app and IDE extension share one configuration file, so adding Connie in any of them adds it to all of them on that machine.

OpenClaw

On the machine running OpenClaw, add Connie with Streamable HTTP and OAuth, then sign in:

openclaw mcp set connie '{"url":"https://mcp.connie.ai","transport":"streamable-http","auth":"oauth","oauth":{"scope":"mcp:full"}}'
openclaw mcp login connie

The login step opens the browser sign-in. Run openclaw mcp doctor connie --probe to check the connection.

Hermes

hermes mcp add --url https://mcp.connie.ai --auth oauth connie

Hermes opens a browser to sign in to Connie. Run hermes mcp test connie to check the connection, then restart the session or run /reload-mcp so the tools load. To sign in again later, run hermes mcp login connie.

Cursor, VS Code, Gemini CLI, Windsurf, Cline and other MCP clients

Any client that supports remote MCP servers over HTTPS can connect. Add a new MCP server, choose the remote, HTTP or "Streamable HTTP" option, paste the server address above, and complete the Connie sign-in when the client asks you to authenticate. Windsurf and Cline use the mcp-remote bridge for the sign-in, with this entry under mcpServers:

{
  "mcpServers": {
    "connie": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.connie.ai"]
    }
  }
}

The Connect my agent page in Connie shows the exact steps for each of these clients.

Claude desktop app

Use Customize → Connectors as described in the Claude section above. The connector is stored with your Claude account, so it is the same connection you see in the browser.

One authorization covers every workspace you are a member of. The workspace you pick on the consent screen becomes the default: any request that does not name a workspace goes there. To work in another workspace, just name it. Ask "List my Connie workspaces" once and the assistant will resolve names to the right workspace from then on. A request for a workspace you are not a member of is refused.

To change the default, revoke the authorization from the Connectors page (see below) and connect again. If you want one client entry locked to a single workspace, append ?workspaceId=<workspace-id> to the server address when you add it; the assistant can still name another workspace explicitly, but anything unspecified stays pinned.

Switching workspaces works with the browser sign-in. A personal API key is tied to the one workspace it was created in.

Connie gives your assistant a small set of tools that open up the rest of your workspace:

1. search-tools finds the right Connie tool for what you asked.
2. get-tool reads what that tool does and what it needs.
3. use-tool runs it.
4. search-skills and get-skill find and read Connie skills, which are step-by-step guides for common tasks.
5. show-video plays a video that a tool or skill returned, in a player in the chat. Only apps that can show one, such as Claude and ChatGPT, list it; elsewhere a video comes back as a link.

use-tool can run any tool in Connie's catalogue, not only the ones your client lists. Together they cover:

• Records and Lists: searching, reading, creating, updating and deleting People, Companies, Deals, Tasks and other Records, and managing Lists.
• CSV import and export into and out of your Lists.
• Lead search and enrichment: finding new people and companies, and adding details such as email addresses, phone numbers and company information to your Records.
• Workflows: building, editing and running them.
• Sending email and messages from the accounts connected to your workspace.
• Connected apps: actions in the apps your workspace has connected, such as your CRM, Slack or outreach tools. These read from and write to those apps, outside your Connie workspace.
• Your workspace: reading workspace details, members, files and skills.

Scope. The connector acts only in the Connie workspaces you are a member of, and every tool runs with your own permissions: a guest in a workspace cannot write through the connector either. Actions in connected apps use the accounts connected to that workspace, and a tool turned off in the workspace's AI tool settings cannot run through the connector.

Credits. Tools that enrich or research consume credits from the workspace's balance exactly as they do in the Connie app. Every result includes a billing object with the credits actually charged, which can differ from an estimate, and every call is attributed to you in the workspace's credit usage log.

Information about people. Lead search and contact enrichment can return personal details about people outside your workspace, such as a name, job title, email address, phone number and professional profile, sourced from third-party business contact data providers and public sources. The results are stored only in your workspace. The Privacy Policy explains where this information comes from, the roles and legal basis, and how someone can ask for it to be removed.

1. In Connie, go to Connectors in the left sidebar and open the MCP view.
2. Under Your MCP clients you see one card per client you have authorized — for example Claude, ChatGPT, Claude Code or Codex — with when it was first granted and last used. Only you can see your own cards; teammates see theirs.
3. Open a card to revoke a single authorization or use Revoke all to disconnect that client everywhere.

Revocation is checked on every request, so the client's next call is refused and it will ask you to sign in again. A sign-in stays valid for as long as the client keeps using it; a client that is idle for 30 days has to sign in again. Access tokens themselves rotate every 8 hours in the background, which you never see.

Removing the connector inside the assistant (for example Remove in Claude's Connectors page or claude mcp remove connie) also ends access from that client, and you can revoke the card in Connie afterwards to tidy up.

The client says "Needs authentication", or calls fail with 401. The sign-in has expired or been revoked. In Claude Code run /mcp and pick Connie, or run claude mcp login connie. In Codex run codex mcp login connie --scopes mcp:full. In Claude or ChatGPT open the connector and choose Connect or Authenticate again.

The client looks connected but its card is missing from Connectors, and calls fail. Connie revokes a client's access as a precaution when it receives a sign-in token that has already been replaced — which happens when two copies of the same sign-in exist, for example when a client was restored from a backup or two instances of it share one stored login. Sign in again from the client; the card reappears on the next successful call. If it keeps happening with Codex, connect the Codex app and the Codex CLI as separate servers, or use only one of them on that machine.

"You are not a member of the requested workspace" (403). The assistant targeted a workspace your account is not in, or you have since been removed from it. Name a workspace you belong to, or ask its admin to add you. Authorization never expands your access.

"This workspace was deleted". The workspace you connected to is in its deletion window. Restore it in Connie, or connect again and pick another default.

A sign-in tab keeps reopening after you revoked access. Some clients retry every few seconds and each retry opens a new consent screen. Remove the Connie entry from the client first (or turn the connector off in the assistant), restart it, and only then revoke in Connie. To reconnect later, add the entry back.

ChatGPT cannot add the connector. Developer mode has to be on, the address must start with https://, and custom connectors are only available on the web, not in the mobile apps.

Claude Team or Enterprise members cannot see "Add custom connector". Only organization owners can add custom connectors; members connect to what the owner has added.

A tool you use in Connie is not available to the assistant. Almost every Connie tool runs through use-tool, even when it is not in your client's list of tools. If get-tool says a tool cannot run through the connector, it gives the reason. A tool turned off in your workspace's AI tool settings is unavailable here too. Tell us which tool you need at the address below if it is still missing.

• Email us at support@connie.ai with the client you are using, the time of the failed request and the message you saw.
• How the connector handles your data, and how long authorization records are kept, is described in the Connie Privacy Policy.
• If you are not a Connie user and want information about you that was found through lead search or enrichment removed, see how to ask for removal.