Skip to main content

​
Overview

mcp-agent cloud deploys your agents as MCP servers, where agent workflows become long-running MCP tools executed in a durable workflow orchestration engine (Temporal). This architecture enables:
  • Agents as MCP Servers: Your agents are exposed as standard MCP servers that any MCP client can connect to
  • Workflows as Durable Tools: Agent workflows are exposed as MCP tools that run durably via Temporal
  • Production-Ready Infrastructure: Built-in retry logic, fault tolerance, and state persistence

​
Key Features

  • Agent Deployment: Deploy agents as MCP servers accessible via HTTP/WebSocket
  • Temporal Integration: Durable workflow execution with automatic retries
  • Secrets Management: Secure storage and injection of API keys
  • Monitoring: Real-time logs, metrics, and workflow tracking

​
Preparing your mcp-agent app as a server

  1. Double check that your entrypoint python file is named main.py
When packaging your mcp-agent app for the cloud, our CLI will be searching for main.py. The entire directory will be deployed, so you can reference other files and upload other assets.
You can optionally exclude files from the bundle using an ignore file (gitignore syntax). Precedence: 1) --ignore-file <path> (explicit override), 2) .mcpacignore in --config-dir, 3) .mcpacignore in the working directory (CWD).
  1. Make sure you have either a pyproject.toml or a requirements.txt file in your app directory.
  2. Mark your functions that you’d like to be tool calls with the @app.tool decorators
main.py
# Your agent definition
@app.tool
async def research(query) -> str:
  # this is where your agent logic exists (functions and calls)
  return results

# Becomes an MCP tool when deployed:
# Tool: "research"

​
Deploying your mcp-agent app

  1. Login into mcp-agent cloud to get your api key
uv run mcp-agent login
  1. Deploy your app
uv run mcp-agent deploy my-first-agent
During the deployment flow, you will be guided through how you’d like your secrets to be managed.

​
Connect to your deployed mcp-agent server

​
Claude Desktop integration

Configure Claude Desktop to access your agent server
.claude-desktop/config.json
"my-agent-server": {
  "command": "/path/to/npx",
  "args": [
    "mcp-remote",
    "https://[your-agent-server-id].deployments.mcp-agent.com/sse",
    "--header",
    "Authorization: Bearer ${BEARER_TOKEN}"
  ],
  "env": {
        "BEARER_TOKEN": "your-mcp-agent-cloud-api-token"
      }
}

​
MCP Inspector

Run your local MCP Inspector 
npx @modelcontextprotocol/inspector
Connect with the following settings
SettingValue
Transport TypeSSE
SSEhttps://[your-agent-server-id].deployments.mcp-agent.com/sse
Header NameAuthorization
Bearer Tokenyour-mcp-agent-cloud-api-token

​
Example Use Cases

Explore use cases in the examples directory:
  • Github to Slack agent: Deploy an agent that will give you and your team a daily report of high-pri PRs and issues from github.
  • Research agent: Deploy a research agent server to do a LLM-based research across the internet.
  • Financial analyzer agent: Deploy a financial analyzer agent to research information about a specific company

​
Next Steps

⌘I