MCP Protocol

Anthropic proposed a novel standard for AI agents to access data sources called the Model Context Protocol (MCP) You can read more about it from Anthropic's blog here. Octusoperator implements MCP to give its operators direct access to your systems and data while maintaining security and control.

How MCP Works in Octusoperator

Copy

from Octusoperator import MCPServer, MCPClient

# Create an MCP server to expose your data
server = MCPServer(
    name="github-connector",
    data_sources=["repos", "issues", "pull_requests"],
    auth_config={"type": "oauth2"}
)

# Connect Octusoperator as an MCP client
client = MCPClient(
    server_url="http://localhost:8000",
    credentials={"access_token": "..."}
)

Key Components

1. MCP Servers: Expose your data sources through a standardized API

• Code repositories (Git, GitHub)

• Documentation (Google Drive, Notion)

• Databases (Postgres, MongoDB)

• Web apps (via Puppeteer)

1. MCP Clients: AI agents that connect to MCP servers

• Octusoperator operators act as MCP clients

• Can access multiple data sources

• Maintain context across interactions

1. Context Management:

• Persistent memory across sessions

• Knowledge graph of relationships

• Semantic search capabilities

Benefits of MCP in Octusoperator

1. Universal Data Access

• Single protocol for all data sources

• No custom integrations needed

• Standardized authentication

1. Context Awareness

• Operators maintain state across calls

• Can reference previous interactions

• Build knowledge over time

1. Security & Control

• Fine-grained access control

• Audit logging

• Rate limiting

Example: GitHub Integration

Copy

# Setup GitHub MCP server
github_server = MCPServer(
    name="github",
    repo="username/repo",
    access_token="..."
)

# Create Octusoperator task with MCP context
task = octus.Task(
    name="code_review",
    mcp_context=[github_server],
    inputs={
        "pr_number": 123,
        "review_type": "security"
    }
)

# Operator can now access GitHub data
result = octus.execute_task(task)

Example: Database Integration

Copy

# Setup Postgres MCP server
db_server = MCPServer(
    name="analytics_db",
    connection_string="postgresql://...",
    allowed_tables=["users", "events"]
)

# Analyze data with MCP context
task = octus.Task(
    name="user_analysis",
    mcp_context=[db_server],
    inputs={
        "metric": "retention",
        "date_range": "last_30_days"
    }
)

result = octus.execute_task(task)

Available MCP Actions

Copy

# Query data source
mcp_query(
    server: str,
    query: str,
    params: Dict = None
) -> Dict

# Update data
mcp_update(
    server: str,
    operation: str,
    data: Dict
) -> None

# Stream changes
mcp_subscribe(
    server: str,
    event_type: str,
    handler: Callable
) -> None

Best Practices

1. Security

• Use minimal access permissions

• Rotate credentials regularly

• Monitor usage patterns

1. Performance

• Cache frequently accessed data

• Use efficient queries

• Implement rate limiting

1. Reliability

• Handle connection errors

• Implement retries

• Monitor server health

Learn More

• MCP Specification

• Building MCP Servers

• MCP Security Guide

• Example Implementations