Back to Help Center
    Developers

    Connecting an AI Assistant via the Teamo MCP Server

    The Teamo MCP server lets AI assistants (Claude Desktop, claude.ai, Claude Code, Cursor, and ChatGPT in developer mode) connect directly to your organization and do real work in it through natural language, instead of you writing API calls yourself.

    This guide walks you through connecting an assistant for the first time and explains what it can and can't do once it's connected.

    Before You Begin

    Before connecting an assistant, ask yourself:

    What do I want it to help with?

    Examples:

    • "Show me overdue records in the Projects module"
    • "Create a Q3 Launch module with status, owner, and deadline columns"
    • "How long did each deal spend in Negotiation?"
    • "Add these five records to the Clients module"

    MCP (Model Context Protocol) is the open standard these assistants use to connect to outside tools. Teamo's MCP server is a thin adapter over the same API described in Using the Teamo API: anything an assistant does through MCP, it could also do through the REST API, with identical results. Because the tool surface tracks that API directly, api.teamo360.com/api/v1/docs is the best place to check for changes as new capabilities ship.

    Step 1: Choose How to Connect

    Server URL: https://api.teamo360.com/api/mcp

    Paste this address when your AI assistant asks for an MCP server URL. In claude.ai, for example, go to Settings → Connectors → Add custom connector.

    There are two ways to connect an assistant:

    Click-and-approve (OAuth)

    From My Profile → Connect an AI Assistant, choose your client and approve access. You'll pick which organization the connection applies to and see exactly what you're granting before you confirm, with no tokens to copy or paste.

    Pasted API token

    For clients that don't support the OAuth flow yet, or for a Claude Code or Cursor setup you want to configure directly, create an API token from My Profile → API Tokens and connect using it instead.

    Examples of when to use a pasted token:

    • Setting up Claude Code with claude mcp add --transport http
    • Configuring Cursor's MCP settings directly
    • A client without OAuth support yet

    For stdio-only MCP clients, install the shim:

    npx @teamo360/mcp

    Step 2: Understand What the Assistant Can Do

    Once connected, the assistant has access to the same capabilities as the API, grouped into tools it can call:

    • Discovery: see your organization's structure: Units, Modules, column schemas, users, and teams
    • Reading: query and search records with filters (including computed and lookup columns), read Activities and Detail rows, pull change history, and read time entries
    • Writing: create, update, and upsert records in batch; archive, delete, and restore; duplicate items; manage Activities and their Detail rows
    • Collaboration: post and read comments, attach and list files
    • Schema: create new Modules, Units, and columns
    • Platform: manage webhook subscriptions

    Deleting a Module or column, and hard deletes anywhere, are intentionally not available to assistants. Those stay UI-only actions.

    Step 3: Know What Stays Protected

    Every connection, whether an OAuth grant or a pasted token, acts as the person who created it, never as a workspace-wide superuser:

    • Your role, module access, and view access apply to every tool call, exactly as they do in the app
    • Columns hidden from you stay hidden from the assistant, including in change history
    • Guest accounts stay sandboxed to their assigned module and view
    • Every action the assistant takes is attributed to you in the audit log and change history
    • You can scope a connection to read-only or a single module, so an assistant only ever gets as much access as it needs

    Writes made by an assistant trigger the same automations, webhooks, and real-time updates as a change made in the UI. Your teammates see it live, and it's never a silent edit.

    Example: Reviewing Overdue Work

    Unit: Operations
    Module: Projects
    Prompt: "Show me anything overdue and assigned to me, then draft a status update comment on each."
    Result: The assistant queries the Projects module filtered by due date and assignee, then posts a comment on each matching record, attributed to you, visible to your team immediately.

    Best Practices

    Start with a scoped or read-only connection

    If you're just exploring what an assistant can do, connect with a read-only or single-module token first. You can always create a broader one once you know what you need.

    Ask the assistant to confirm before deleting anything

    Deletes require an explicit confirmation step by design. If an assistant asks before deleting, that's expected, not a bug.

    Review what changed

    Because every assistant action lands in your change history under your name, it's easy to review afterward. Treat it the same way you'd review a teammate's edits.

    Teamo Tip: If an assistant seems unsure which record or column you mean, it's usually because your workspace has more than one close match. Being specific (module name, exact column label) gets a faster, more accurate result.

    Next Steps

    Now that you've connected an assistant, continue with: