Skip to main content
Version: Cloud

MCP Server

The Deeploy Model Context Protocol (MCP) server connects any MCP-compatible AI agent platform to Deeploy. Once connected, an AI agent registers itself with Deeploy and automatically sends an audit trail of every interaction — without any code changes to the agent.

The same server assists compliance officers with use case controls: reviewing controls, drafting evidence, and linking documents. That workflow is manual and opt-in, so it never runs unless a user asks for it.

Availability​

The MCP server is hosted on Deeploy cloud and available to all cloud customers.

Private cloud customers can also run the server in their own environment, and can additionally enable login with their own identity provider (SSO) so users sign in with their corporate account instead of a Deeploy credential. Both are configured by an administrator in the Helm chart — see Identity provider login. Contact support@deeploy.ml for installation instructions.

What the server gives you​

CapabilityAutomatic?Who it is for
Agent registration and tracingYes — runs at the start of every sessionAll users
Use case controlsNo — you ask for it explicitlyCompliance officers
Skills over MCPGuided by instructionsAll users / compliance officers

These capabilities are deliberately separate. Registration and tracing are automatic and affect everyone. Controls work is manual, so an agent never starts it unprompted — not all users are compliance officers.

Supported platforms​

Any MCP-compatible AI platform can connect. Tested platforms include:

  • Microsoft Copilot Studio
  • Claude (claude.ai) / Claude Desktop / Claude Code
  • ChatGPT
  • GitHub Copilot
  • Perplexity

Connect your platform​

The MCP server URL is:

  • Deeploy cloud — https://mcp.app.deeploy.ml/mcp
  • Private cloud — https://mcp.<base_url>/mcp, where <base_url> is your Deeploy host

Add it as a custom connector in your platform, then:

  1. Authenticate when prompted. On cloud deployments you supply your Deeploy oauth credentials (client_id and client_secret). Private cloud admin may configure an identity provider sign in — see Identity provider login.
  2. Add a system instruction telling the agent to use the Deeploy MCP server at the start of every conversation, so the flow runs automatically rather than needing a manual prompt each time. See System instruction below.
  3. Once connected, the agent automatically follows the registration flow at the start of every session.

The server communicates using Streamable HTTP and authenticates over OAuth. Your platform handles the handshake; the sections below cover the platform-specific steps.

Claude​

Add a custom connector with the URL above. See the Claude connectors documentation for the exact steps.

GitHub Copilot​

Add the server to an mcp.json file, or run MCP: Add Server from the Command Palette:

{
"servers": {
"deeploy": {
"type": "http",
"url": "https://mcp.app.deeploy.ml/mcp"
}
}
}

See the VS Code MCP documentation for other ways to add a server and for configuring it in other Copilot surfaces.

Microsoft Copilot Studio​

Add the URL above as a Model Context Protocol tool on your agent. See the Copilot Studio MCP documentation for the exact steps.

ChatGPT​

Add a connector with the URL above. See Connectors in ChatGPT for the exact steps.

Perplexity​

Add a custom connector with the URL above. See the Perplexity MCP documentation for the exact steps.

Other MCP clients​

Any MCP-compatible platform can connect using the same URL over Streamable HTTP. Consult your platform's documentation for how to add a remote, OAuth-authenticated MCP server.

System instruction​

The MCP server instructs the agent to start the registration flow at the beginning of every session. To make sure this happens automatically, without you needing to prompt the agent manually, add a system instruction that tells the agent to always use the Deeploy MCP server at the start of each conversation.

You can add a system instruction at the organization level (applies to all users) or at the personal level (applies to your own sessions only). Refer to your platform's documentation for the exact steps.

Add the following text as the system instruction:

At the start of every conversation, use the Deeploy MCP server to register this agent session and trace all interactions. The Deeploy MCP connector can be configured at organization level, user will still need to connect and authenticate individually.

Adding this instruction is what makes registration and tracing reliable. Without it, the agent may decide to skip registration.

Registration flow​

At the start of every session, the agent automatically follows a three-step flow:

Step 1 — List agents. The agent fetches all agents registered in your Workspace and presents them to you. If you have multiple Workspaces, you're asked to select one first.

Step 2 — Register. The agent asks you to choose a name. It then registers (or looks up) that name in Deeploy. Registration is idempotent — using the same name again adds a new session trace to the existing registration instead of creating a duplicate.

Step 3 — Trace all interactions. Once registered, the agent silently sends an audit trail entry to Deeploy after every response for the rest of the session. No action is required on your end.

Type skip at step 2 to continue the session without registering or tracing.

Adding the .deeploy-autoregister skill makes this flow more robust. See Skills over MCP.

Where to go next​