AI Agent Access (MCP Server)

The Sovereign MCP Server lets external AI agents — like Claude, GPT, or your own custom agents — connect to your deployment and use your workflows as tools. Your AI workforce can trigger automations, query data, and manage workflows through a standard protocol.

What Is MCP?

MCP (Model Context Protocol) is a standard for connecting AI agents to external tools and data sources. When your Sovereign deployment runs the MCP Server, any MCP-compatible AI agent can discover and invoke your workflows as if they were native tools.

This means your AI agents do not need custom integration code for each automation — they connect once to Sovereign and gain access to everything your workflows can do.

How It Works

1. Connect — point your AI agent at the MCP Server URL on your deployment
2. Authenticate — the agent goes through OAuth 2.1 authentication (the same standard used by all modern services)
3. Discover — the agent automatically discovers all available tools, including your workflow actions and platform tools
4. Use — the agent can invoke any tool it discovers, just as it would use any other tool

The MCP Server exposes your registered connector actions as tools that agents can invoke. It also provides platform-native tools for managing the automation system itself.

Platform Tools

In addition to connector actions, the MCP Server provides built-in platform tools:

These tools let an AI agent not only execute your existing automations but also create and manage new ones.

Search and Discovery

When your deployment has many connector actions available, the MCP Server provides a search gateway that lets agents efficiently find the right tools for their task. Rather than receiving hundreds of tool definitions up front, agents can search for relevant tools by keyword and category.

Connecting an AI Agent

Access URLs

The MCP Server is accessible through the reverse proxy or directly:

Authentication Flow

MCP-compatible agents handle the OAuth 2.1 flow automatically. The process is:

1. Agent connects to the MCP Server URL
2. Server returns authentication requirements
3. Agent discovers the Auth Service via standard OAuth metadata endpoints
4. Agent registers itself as a client (Dynamic Client Registration)
5. Agent goes through the authorization flow (with PKCE for security)
6. After authorization, the agent receives an access token

For AI agents that support MCP natively (like Claude), this flow is handled automatically — you just provide the URL.

Session Modes

The MCP Server supports two session modes:

• Interactive — multi-turn conversation where the agent sends messages, receives responses, and calls tools in a back-and-forth dialogue
• Bounded — single-turn execution with budget limits on iterations, tool calls, tokens, and duration. Used for automated workflow step execution.

Budget controls prevent runaway sessions. When any configured limit is reached, the session terminates cleanly.

Setting Up MCP Access

The MCP Server is included in the standard Sovereign Workflows deployment. To enable AI agent access:

1. Ensure the MCP Server is running (it is included in the Docker Compose configuration)
2. Verify the MCP Server is accessible at its configured URL
3. If using the reverse proxy, the MCP Server is available at the /mcp/ path
4. Configure LLM provider API keys if you want agents to have AI reasoning capabilities
5. Point your AI agent at the URL and complete the authentication flow

Next Steps

• Autonomous Agent Nodes — use AI agents within your workflows
• AI-Powered Workflows — add AI steps to workflows
• Authentication Setup — configure the OAuth 2.1 authentication layer