MCP Overview

Introduction

MCP (Model Context Protocol) is a standard for connecting AI models to external tools and data sources. Greentic implements MCP for:

• Tool execution in sandboxed WASM environment
• Secure external API access
• Database queries
• File system operations

Key Features

WASM Sandboxing

MCP tools run in isolated WebAssembly environments for security.

Tool Discovery

Automatic tool registration and capability advertisement.

Type Safety

WIT interfaces ensure type-safe tool invocations.

Observability

Built-in tracing and logging for tool executions.

Architecture

LLM (with tool use)
       │
       ▼ Tool Call Request
┌─────────────────────────────────┐
│       greentic-mcp              │
│   (MCP Executor / WASI Bridge)  │
└─────────────────────────────────┘
       │
       ▼
┌─────────────────────────────────┐
│    WASM Tool Component          │
│   (Sandboxed execution)         │
└─────────────────────────────────┘
       │
       ▼ Tool Result
Back to LLM

MCP in Flows

Tool Definition

tools/database-query.yaml

name: database_query
description: Query the customer database
parameters:
  - name: query
    type: string
    description: SQL query to execute
    required: true
  - name: limit
    type: integer
    description: Maximum rows to return
    default: 10
returns:
  type: array
  items:
    type: object

Using Tools in Flow

- id: query_orders
  type: mcp-tool
  config:
    tool: "database_query"
    parameters:
      query: "SELECT * FROM orders WHERE customer_id = '{{customer_id}}'"
      limit: 5
  output: orders_result
  next: process_orders

With LLM

- id: agent_step
  type: llm
  config:
    model: "gpt-4"
    system_prompt: "You are a helpful assistant with database access."
    tools:
      - database_query
      - send_email
      - create_ticket
    tool_choice: "auto"
  next: handle_response

Available Tools

Greentic provides several built-in MCP tools:

Tool	Description
http_request	Make HTTP requests
database_query	Query databases
file_read	Read files (sandboxed)
file_write	Write files (sandboxed)
send_email	Send emails
create_ticket	Create support tickets

Security

WASM Sandboxing

All MCP tools run in WASM sandboxes:

• No direct filesystem access
• No network access unless explicitly granted
• Memory isolation
• CPU time limits

Capability Model

Tools declare required capabilities:

capabilities:
  - network:outbound  # Can make outbound requests
  - filesystem:read   # Can read files in sandbox

Next Steps

• Creating MCP Tools
• LLM OpenAI Component