FastMCP Migration & Comparison

mcp-fastmcp-instructions.md

Project: FastMCP Migration & Comparison

Objective
Migrate your existing FastAPI services to the Model Context Protocol (MCP) using two different implementations:

• Official MCP SDK (mcp on PyPI)
• Jeremiah Lowin’s FastMCP 2.0 (fastmcp on PyPI)

Evaluate each on performance, LangChain integration, and production readiness.

---

Critical Inputs

1. Existing FastAPI Code (REPLACE WITH YOUR REAL CODE):

   from fastapi import FastAPI
   app = FastAPI()
   
   @app.post("/query")
   async def query_engine(text: str):
       # Your actual retrieval/RAG logic here
       return {"response": f"Processed {text}"}

2. Project Constraints:
   • Scale: e.g. 5 endpoints, 100+ QPS
   • Non-negotiables: e.g. OAuth2, latency <8 ms, Kubernetes
   • LangChain Usage: e.g. LCEL chain with OpenAI + pgvector
   • Baseline Metric: e.g. Current latency: 15 ms @ 50 QPS
3. Environment:
   • Python version (e.g. 3.11)
   • FastAPI version (e.g. 0.98)
   • OS/container details if relevant

---

Response Requirements

A. Code Conversion

• Produce side-by-side FastAPI → FastMCP examples for your endpoint, including:
  • ASGI mounting & lifespan events
  • @tool definitions
  • Full runnable snippets with robust error handling (retries, circuit-breakers)

B. Architecture Analysis

Comparison Point	Official MCP SDK (mcp)	FastMCP 2.0 (fastmcp)
Cold start	[benchmark vs. baseline]	[benchmark vs. baseline]
Async throughput	[req/s @ 50 concurrent]	[req/s @ 50 concurrent]
License Risk	Apache 2.0 (✅)	MIT (✅)
PyPI Downloads (30 d)	[data]	[data]

Evidence Rule: Cite PyPI/GitHub stats; if unavailable, explain how to measure (e.g. locust or k6).

C. LangChain Integration

• Show how to wrap your LangChain chain using langchain-mcp-adapters
• Provide a compatibility matrix for both MCP implementations

D. Production Practices

# IMPLEMENT IN CODE:
# 1. OAuth2 middleware ▶ WHY: SSO compliance
# 2. OpenTelemetry hooks          ▶ WHY: distributed tracing
# 3. Async resource cleanup      ▶ WHY: prevent connection leaks
# 4. Retry & circuit-breaker     ▶ WHY: resilience under load

E. Decision & Rollback

• Final recommendation with one-sentence justification
• 3-step rollback plan to revert to FastAPI

---

Tone & Format

• Max 1,200 words
• Prioritize: Code > Tables > Bullet lists
• Annotate every code snippet with ▶ WHY comments
• Zero fluff—focus on actionable, evidence-backed guidance

---

Fallback Protocol

Missing benchmarks? Provide ready-to-run locust or k6 scripts to gather the required metrics.

# Example locust task
from locust import FastHttpUser, task, between

class TestUser(FastHttpUser):
    wait_time = between(1, 2)

    @task
    def query(self):
        self.client.post("/query", json={"text": "test"}, name="/query")

# Example k6 script
import http from 'k6/http';
import { sleep } from 'k6s';

export default function () {
  http.post('https://your-host/query', JSON.stringify({ text: 'test' }), { headers: { 'Content-Type': 'application/json' } });
  sleep(1);
}