Use the agent you already trust

ConvergeSense exposes reviewed, workspace-scoped infrastructure tools over OAuth-protected Streamable HTTP MCP. Device credentials and raw sessions stay on the onsite Engine.

Before connecting#

An owner needs to enroll an Engine, activate a reviewed connection, and enable the intended tools for your role and client. The Engine must advertise agent_tool_plane_v1.

Remote MCP endpoint#

Configure your client with:

https://mcp.convergesense.com/mcp

Your client discovers OAuth metadata automatically, opens ConvergeSense sign-in and consent, and binds the grant to one workspace. If you belong to multiple workspaces, choose the one whose systems you intend to use.

Supported clients#

ChatGPT and Codex#

Use the ConvergeSense Agent plugin when available to you, or add the remote MCP endpoint through your client's supported connection flow. Complete the browser OAuth sign-in and consent. Availability of custom integrations depends on your client and account settings.

Claude and Claude Code#

Add the endpoint as a remote MCP integration. Complete OAuth in the browser, choose your ConvergeSense workspace, and review the requested permissions.

Microsoft Copilot Studio#

Add an existing MCP server using Streamable HTTP and the endpoint above. Use the discovered OAuth metadata. Dynamic Client Registration is supported.

VS Code, Cursor, and other clients#

Use a client that supports remote Streamable HTTP MCP and OAuth. Enter the endpoint in its setup flow. Configuration file formats vary by client; this is the servers-style configuration used by compatible clients such as VS Code:

{
  "servers": {
    "convergesense": {
      "type": "http",
      "url": "https://mcp.convergesense.com/mcp"
    }
  }
}

If your client expects a different root key, follow its own configuration format and use the same endpoint. Do not put a device credential in the client configuration.

Verify the connection#

Ask the agent which workspace it is connected to and which approved tools are available. The normal sequence is convergesense_get_context, then search, describe, and run an available tool.

Long-running reads return a job ID promptly. Your agent polls convergesense_get_job until the result is ready. See the MCP reference.

Permissions and revocation#

Viewer, operator, and owner permissions come from your account and workspace tool policy. An OAuth grant never expands your application role.

Owners can disable or revoke an integration from Agent connections. Removing a member also revokes that user's active agent grants and sessions.

Data-sharing notice#

When you invoke a tool, its inputs and returned infrastructure or customer context are sent to your selected AI provider. That provider's subscription, retention, and training terms apply to its copy. ConvergeSense does not store the external agent conversation.

Read Security and data before connecting an agent to sensitive systems.