Step 2: Exposing Your Crew via API

Now that we have a working CrewAI service, the next step is to expose it via an API so external users can interact with it. We'll use FastAPI, a lightweight and high-performance web framework for Python.

This API allows users to: - Start a new AI task and create a payment request - Check the job and payment status

1. Look around the example API

Now, if you have modified main.py in the previous steps, return it to the initial state.

Note how API is configured in main.py

In order to work with Masumi, API must follow MIP-003 (the Masumi Protocol Standard) and include all the endpoints required by the standard.

Since we only have an example agent and the actual business logic may differ, you will see placeholder logic in some endpoints (e.g., generating payment_id, handling partial states, and provide_input)

Temporary job storage - not for production!

For simplicity, we store jobs in a Python dictionary (jobs = {}).

This has serious limitations - jobs will be lost if the server restarts.

Do not use this in production!

In a production environment, you should:

1. Store jobs in a database (e.g., PostgreSQL, MongoDB, Redis).

2. Possibly integrate a message queue system (e.g., RabbitMQ, Celery, Kafka) for background job processing.

3. Running the API

Run the FastAPI server:

python main.py api

The API Swagger will be available at:

http://localhost:8000/docs

This automatically provides interactive documentation for all endpoints.

4. Testing the MIP-003 Endpoints

Copy & paste the following cURL commands to your Terminal to test each endpoint.

Retrieving the Input Schema (GET /input_schema)

cURL:

curl -X curl -X GET "http://localhost:8000/input_schema"

Example Response:

{
"input_data":[
    {
        "id":"text",
        "type":"string",
        "name":"Task Description",
        "data":{
            "description":"The text input for the AI task",
            "placeholder":"Enter your task description here"
        }
    }
]
}      

Checking Server Availability (GET /availability)

cURL:

curl -X GET "http://localhost:8000/availability"

Example Response:

{
    "status":"available",
    "type":"masumi-agent",
    "message":"Server operational."
}

Starting a Job (POST /start_job)

This will fail if the Masumi payment service not yet running (We will configure it in Step 3)

cURL:

curl -X POST "http://localhost:8000/start_job" \
-H "Content-Type: application/json" \
-d '{
        "identifier_from_purchaser": "example_purchaser_123", 
        "input_data": {"text": "coffee beans"}
        }'

Example Response:

{
  "status": "success",
  "job_id": "e4a1f88c-769a-4e8d-b298-3b64345afa3b",
  "payment_id": "abcd-efgh-5678"
}

Checking Job Status (GET /status?job_id=<job_id>)

cURL:

curl -X GET "http://localhost:8000/status?job_id=<job_id>"

Example Response (Pending or Completed):

{
  "job_id": "e4a1f88c-769a-4e8d-b298-3b64345afa3b",
  "status": "completed",
  "result": "Processed input data: {'text': 'Research about AI governance models', 'option': 'summary'}"
}

Response If the job ID doesn’t exist:

{
  "error": "Job not found"
}

Providing Additional Input (POST /provide_input)

cURL:

curl -X POST "http://localhost:8000/provide_input" \
     -H "Content-Type: application/json" \
     -d '{
           "job_id": "<job_id>",
           "input_data": [
             { "key": "text", "value": "example text" }
           ]
         }'

Example Response:

{
  "status": "success"
}

---

6. Summary

✅ Your CrewAI service now implements the MIP-003 Standard for Agentic Services.

✅All crucial endpoints (/start_job, /status, /provide_input, /availability, /input_schema) are exposed.

✅Jobs are tracked in memory for this tutorial—not recommended for production. In a real production environment, consider using a reliable database and/or a queue system.

---