MCP & Claude

How On Belay connects to Claude Desktop and claude.ai, and a reference for every MCP tool On Belay exposes.

What is MCP

MCP (Model Context Protocol) is an open standard that lets Claude call external tools and retrieve structured context during a conversation. On Belay implements an MCP server at https://app.onbelay.ai/api/mcp.

When Claude connects to On Belay via MCP, it gains access to four tools that surface your organization's context — who you are, what groups you're in, what integrations you can use, and how to authenticate against them.

Authentication uses OAuth 2.0 with PKCE. No tokens are embedded in config files or shared manually — the flow is handled automatically when you add the server.

Claude Desktop setup

Open Claude Desktop and navigate to Settings → Developer → MCP Servers. Add the following server configuration to your claude_desktop_config.json:

{
  "mcpServers": {
    "onbelay": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/inspector"],
      "env": {
        "SERVER_URL": "https://app.onbelay.ai/api/mcp",
        "BEARER_TOKEN": "<your-token>"
      }
    }
  }
}

Replace <your-token> with your On Belay bearer token. Find your token in Dashboard → Settings → Claude connection or complete the OAuth flow from the settings page to generate one.

After saving: Restart Claude Desktop. You'll see "onbelay" appear in the MCP tools panel. The four On Belay tools will be available in all conversations.

claude.ai setup

claude.ai supports MCP servers natively. No config file needed — it's a UI flow.

  1. 1

    Go to claude.ai/customize/connectors.

  2. 2

    Click the + button to add a new connector.

  3. 3

    Name it On Belay and paste the URL: https://app.onbelay.ai/api/mcp

  4. 4

    Click Connect. You'll be redirected to On Belay's OAuth flow — sign in with your On Belay account to authorize.

  5. 5

    Once authorized, On Belay tools will be listed under the connector and available in every new conversation.

MCP tool reference

get_my_context()

Returns the calling user's full context: their groups, each group's Claude role, people guidelines, and any standing instructions.

Returns: Object with groups array, each containing name, claudeRole, and guidelines fields.
When to call: Call this at the start of a session to load the user's organizational context into Claude.
get_group_context(group_name: string)

Returns the full context for a specific group by name. Useful when a user needs to reference another team's setup.

Returns: Object with group name, Claude role, guidelines, and member count.
When to call: Call this when a task requires understanding a specific team's context — e.g. 'What does the Marketing group focus on?'
list_my_integrations()

Returns all integrations accessible to the user via their group memberships, including permitted scopes for each.

Returns: Array of integrations with slug, name, category, authType, and permitted scopes.
When to call: Call this before attempting any integration action to confirm access and understand what scopes are available.
get_integration_credential(slug: string)

Returns the decrypted credential for a specific integration the user has access to. Only available for integrations in the user's permitted list.

Returns: Object with apiKey or OAuth token, depending on the integration's auth type.
When to call: Call this when you need to authenticate against a specific integration's API on behalf of the user.

Having trouble connecting?

We can help debug MCP setup issues for Claude Desktop or claude.ai.

Contact support