@a5c-ai/aeq-mcp-tool

MCP (Model Context Protocol) tool server for the Ask Expert Question system. Routes questions from AI agents to domain experts instead of asking the current user directly.

Installation

Run directly with npx:

npx @a5c-ai/aeq-mcp-tool

Or install globally:

npm install -g @a5c-ai/aeq-mcp-tool

Configuration for Claude Code

Add the following to your .mcp.json file:

{
  "mcpServers": {
    "ask-expert-question": {
      "command": "node",
      "args": ["packages/mcp-tool/dist/index.js"],
      "env": {
        "AEQ_SERVER_URL": "https://aeq.a5c.ai/api/v1",
        "AEQ_AUTH_TOKEN": "your-token-here"
      }
    }
  }
}

Configuration for Codex

Add the following to .codex/config.toml in the project:

[mcp_servers.ask-expert-question]
command = "node"
args = ["packages/mcp-tool/dist/index.js"]
env = { AEQ_SERVER_URL = "https://aeq.a5c.ai/api/v1" }

Environment Variables

| Variable | Description | Default | |----------|-------------|---------| | AEQ_SERVER_URL | URL of the AEQ server | https://aeq.a5c.ai/api/v1 | | AEQ_AUTH_TOKEN | Authentication token for the AEQ server | (none) |

Available Tools

ask_expert_question

Submit a question to a domain expert. The question is routed to the most appropriate expert based on the domain and content. Returns a question ID for tracking.

check_question_status

Check the status of a previously submitted question. Returns the current state (pending, answered, expired) and the answer if available.

cancel_expert_question

Cancel a pending question that is no longer needed. Removes the question from the queue so experts are not prompted unnecessarily.

list_domain_experts

List all available domain experts and their areas of expertise. Useful for understanding which domains are covered before submitting a question.

How It Works

1. An AI agent calls ask_expert_question via MCP with a question and optional domain hint.
2. The MCP tool forwards the question to the AEQ server through the SDK.
3. A domain expert picks up the question via the CLI or SDK and provides an answer.
4. The AI agent retrieves the answer using check_question_status or receives it via polling.

License

MIT