Agents

These guides show how to connect popular AI coding assistants to Maybe Don’t. Each example uses GitHub’s MCP server as the downstream service — it’s a practical example since most developers already have a GitHub account. Substitute any MCP server you like; the agent-side wiring is the same.

Prerequisites

Before following any of these guides, you’ll need:

1. Maybe Don’t running — Follow Get Started first
2. A downstream MCP server — The examples use GitHub’s MCP server, so you’ll need a GitHub Personal Access Token. If you’re using a different MCP server, substitute its URL and auth configuration
3. Your AI coding assistant installed — Links provided in each guide

Choose Your Agent

Common Pattern

All these guides follow the same pattern:

1. Configure the gateway with a downstream MCP server (GitHub in these examples)
2. Start the gateway listening on HTTP
3. Configure your AI agent to connect to the gateway instead of the MCP server directly
4. Verify the connection works

The gateway configuration is identical across all agents. Only the agent-side configuration differs.

Gateway Configuration

Every example uses this gateway configuration:

server:
  type: http
  listen_addr: "0.0.0.0:8080"

downstream_mcp_servers:
  github:
    type: http
    url: "https://api.githubcopilot.com/mcp/"
    auth:
      pass_through:
        enabled: true
        headers:
          - source_header: "X-GitHub-Token"
            target_header: "Authorization"
            format: "Bearer {value}"

validation:
  ai:
    provider: openai
    endpoint: "https://api.openai.com/v1/chat/completions"
    model: "gpt-4o-mini"
    api_key: "${OPENAI_API_KEY}"

request_validation:
  cel:
    enabled: true
    mode: audit_only
    rules_file: "cel_request_rules.yaml"
  ai:
    enabled: true
    mode: audit_only
    rules_file: "ai_request_rules.yaml"

Save this as config/maybe-dont.yaml and start with:

Now follow the guide for your specific AI agent.