Connect any agent

Set it up

Any MCP client

Most clients take a JSON config. Add the server under its MCP servers:

{
  "mcpServers": {
    "probed": {
      "url": "https://mcp.probed.chat"
    }
  }
}

Codex

codex mcp add probed --url https://mcp.probed.chat
codex mcp login probed

Claude Code

claude mcp add --transport http probed https://mcp.probed.chat

Claude and other chat apps

Look for custom connectors in settings (in Claude, Settings → Connectors → Add custom connector) and paste the server URL above.

Cursor and VS Code

Cursor uses the JSON above in ~/.cursor/mcp.json. VS Code uses .vscode/mcp.json:

{
  "servers": {
    "probed": {
      "type": "http",
      "url": "https://mcp.probed.chat"
    }
  }
}

Approve it

Your client asks you to sign in once: run codex mcp login probed in Codex, /mcp then Authenticate in Claude Code, or click Connect in most apps. A browser page opens the Probed desktop app, which asks you to allow it. Click Allow and the client stays signed in.

To disconnect an agent, open Settings → Developer in Probed and revoke it under Connected integrations.

Without the desktop app

For CI, servers or any machine that can't open the Probed app, use a service token instead. In Probed, open Settings → Developer and, under Service tokens, choose Create token. Copy it (it starts with prbd_st_ and is shown once), then send it as a header:

{
  "mcpServers": {
    "probed": {
      "url": "https://mcp.probed.chat",
      "headers": {
        "Authorization": "Bearer prbd_st_…"
      }
    }
  }
}

In Codex, keep the token in an environment variable:

export PROBED_TOKEN=prbd_st_…
codex mcp add probed --url https://mcp.probed.chat --bearer-token-env-var PROBED_TOKEN

In Claude Code:

claude mcp add --transport http probed https://mcp.probed.chat \
  --header "Authorization: Bearer prbd_st_…"

Treat it like a password. Revoking it in Settings → Developer stops every client using it straight away.

What it can do

You don't need to call tools by name; ask your agent in plain words and it picks the right one. For reference:

Chat

• list_channels, read_recent and search_messages read the channels you can see.
• post_message posts as you, in a channel or as a thread reply.
• manage_message edits or deletes your own messages, reacts and pins.

Space and members

• get_space_plan and get_community describe the space and what its plan includes.
• list_members, manage_members, list_roles and manage_roles look after people and roles.
• create_channel, update_channel and manage_channel_archive organise channels.
• create_invite and manage_invites invite people by email and revoke invites.
• moderate handles timeouts, kicks, bans, message cleanup and slow mode.

Feedback and roadmap

• list_ideas, triage_idea, accept_idea and merge_idea work through the ideas board.
• manage_roadmap, manage_events and manage_feedback run the roadmap, events and feedback forms.

Customers

• manage_customers, manage_crm_notes, manage_canned and manage_badges cover customer workspaces, tags, notes, canned replies and badges.

Housekeeping

• whoami checks who the agent is signed in as.
• revoke_token lets an agent revoke its own service token when a job ends.

Permissions and plans

• The agent acts as you. Every call checks your permissions in that space, so it can't do anything you couldn't do in the app.
• It needs a space on Premium or Business. Spaces on the Free plan don't show up, and neither do spaces where you're a customer. If none of your spaces are on a paid plan, the server won't connect.
• Settings → Developer, where you create service tokens and disconnect agents, is shown to space admins.
• Everything it does is attributed to you, and moderation actions land in the space's moderation log.

MCP or API?

Use the MCP server when an AI agent works for you, as you. Use the Probed API when your own code needs to read or automate your space (onboarding, offboarding, roles, channels, moderation), with a key that belongs to the space rather than to a person.