MCP Agent Governance

What is MCP?

The Model Context Protocol (MCP) is a standard for connecting AI agents to tools and data sources. GateFlow's MCP implementation adds governance, permissions, and observability for production agent deployments.

The Challenge

AI agents are powerful but risky in production:

• Unrestricted Access: Agents can call any tool, access any data
• No Audit Trail: Hard to track what agents did and why
• Cost Surprises: Runaway agents can burn through API budgets
• Compliance Gaps: No way to enforce data handling policies

GateFlow's Solution

GateFlow MCP provides governed agent infrastructure:

Key Features

Default-Deny Permissions

Agents start with zero permissions. You explicitly grant what they need:

agent:
  name: "Support Bot"
  permissions:
    tools:
      - llm/chat           # Can use LLM
      - retrieval/search   # Can search knowledge base
      # Cannot: transcribe audio, synthesize speech, etc.
    models:
      - gpt-4o-mini        # Only this model
    data_classification:
      - public             # Only public data

Learn more →

Complete Audit Trail

Every tool invocation is logged:

{
  "timestamp": "2024-01-15T10:30:00Z",
  "agent_id": "agent_support_bot",
  "tool": "retrieval/search",
  "input": {"query": "password reset"},
  "output": {"results": 5},
  "latency_ms": 234,
  "cost": 0.0012,
  "session_id": "sess_abc123"
}

Learn more →

Cost Transparency

Real-time cost tracking per agent:

{
  "agent_id": "agent_support_bot",
  "session": {
    "cost": 0.45,
    "tools_called": 23,
    "duration_seconds": 180
  },
  "limits": {
    "cost_per_session": 5.00,
    "cost_daily": 100.00
  }
}

Learn more →

Tool Catalog

GateFlow provides governed access to:

Category	Tools	Description
Voice	transcribe, synthesize, pipeline	Audio processing
Documents	ocr, process, status	Document handling
Retrieval	search, rerank	Knowledge base access
LLM	chat, embed	AI model access
Self-Inspect	whoami, get_my_usage	Agent introspection

Explore tools →

Pipeline Templates

Pre-configured pipelines for common use cases:

Template	Description
ambient-scribe	Medical transcription with PII handling
voice-agent-fast	Low-latency voice assistant
voice-agent-premium	High-quality voice assistant
legal-dictation	Legal document transcription

Browse templates →

Quick Start

1. Create an Agent

curl -X POST https://api.gateflow.ai/v1/mcp/agents \
  -H "Authorization: Bearer gw_prod_..." \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Support Bot",
    "permissions": {
      "tools": ["llm/chat", "retrieval/search"],
      "models": ["gpt-5-mini"]
    }
  }'

Response:

{
  "agent_id": "agent_abc123",
  "api_key": "gf-agent-xyz789...",
  "mcp_endpoint": "https://mcp.gateflow.ai/agent_abc123"
}

2. Connect Your Agent

from langchain.tools import MCPToolkit

toolkit = MCPToolkit(
    server_url="https://mcp.gateflow.ai/agent_abc123",
    api_key="gf-agent-xyz789..."
)

tools = toolkit.get_tools()

3. Use Tools

# Agent can now use permitted tools
result = tools["retrieval_search"].invoke({"query": "password reset"})

Full quickstart →

Architecture

Transport Options

Transport	Use Case
HTTP/SSE	Cloud agents, serverless
stdio	Local agents, development

Authentication

Agents authenticate with gf-agent-* prefixed keys:

Authorization: Bearer gf-agent-xyz789...

Session Management

Each agent session is isolated:

• Separate audit logs
• Independent cost tracking
• Session-scoped context

Compliance

GateFlow MCP supports:

• HIPAA: PHI handling with BAA
• SOC 2: Audit logging, access controls
• EU AI Act: Human oversight, risk classification
• GDPR: Data residency, deletion rights

Learn more →

Next Steps

• Create Your First Agent - Hands-on tutorial
• Agent Templates - Pre-built configurations
• Permissions Deep Dive - Security model
• Tool Catalog - Available tools