[BUG] The fatal message: Interrupted · What should Claude do instead?

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When sent a promt to claude code to create a MCP server and client with a function to connect to a SQL DB service to check where some clients spend more than X euros in an order. The session suddenly after try to fetch some info from Web stop with this message:

Interrupted · What should Claude do instead?

Then I send another promt like:

Continue with the loop, and the session finalixed correctly with my code created in my codebase

Version Claude Code: 2.1.227 (Claude Code)

What Should Happen?

Start and finalize the session without any message like: Interrupted · What should Claude do instead?

Error Messages/Logs

Interrupted · What should Claude do instead?

Steps to Reproduce

This is my prompt:

Build an MCP (Model Context Protocol) server and a matching test client using

FastMCP (Python, https://github.com/jlowin/fastmcp), with a tool that
connects
to a PostgreSQL database and retrieves a customer's orders above a given
amount
in euros.

PROJECT STRUCTURE:

• mcp_server.py — FastMCP server exposing the tool
• mcp_client.py — simple test client that connects to the server and calls

the tool

• .env.example — template for DB connection env vars (no real credentials)
• requirements.txt — fastmcp, asyncpg (or psycopg2-binary), python-dotenv
• README.md — setup and run instructions

DATABASE SCHEMA (assume this structure, adjust if I provide a real one
later):

• customers table: id (PK), name, email
• orders table: id (PK), customer_id (FK -> customers.id), order_date,

total_amount (numeric, in euros), status

MCP SERVER REQUIREMENTS:

1. Use FastMCP's decorator-based tool definition (@mcp.tool()).
2. Tool name: get_customer_orders_above_amount

• Parameters: customer_id (int) OR customer_email (str, optional

alternative
lookup), min_amount (float, euros)

• Returns: list of orders (order_id, order_date, total_amount, status)

matching customer_id/email AND total_amount >= min_amount, ordered by
order_date descending

1. Use an async PostgreSQL driver (asyncpg preferred) with a connection pool

created once at server startup, not per-request.

1. Read DB connection params (host, port, dbname, user, password) from

environment variables via python-dotenv — never hardcode credentials.

1. Use parameterized queries only (no string formatting into SQL) to prevent

SQL injection.

1. Handle and return clear errors for: customer not found, invalid amount,

DB connection failure — do not let raw exceptions/stack traces leak to
the client.

1. Add a docstring to the tool function describing its purpose, parameters,

and return format, since FastMCP uses docstrings for tool discovery/schema.

1. Run the server using stdio transport by default (standard for local MCP

servers), but structure the code so switching to HTTP/SSE transport later
is straightforward.

MCP CLIENT REQUIREMENTS:

1. Use FastMCP's client library to connect to the server via stdio.
2. Demonstrate calling get_customer_orders_above_amount with example

parameters (a sample customer_id and min_amount, e.g. 100.0).

1. Print the results in a readable format (table or formatted list).
2. Include basic error handling for connection failures.

ADDITIONAL NOTES:

• Follow FastMCP's current API (fastmcp v2.x) — check import paths carefully,

since fastmcp.server.FastMCP is the current pattern, not older mcp.server
patterns.

• Include type hints throughout.
• Keep the connection pool lifecycle tied to the server's lifespan/startup-

shutdown hooks if FastMCP supports that pattern, rather than opening
connections ad-hoc.

• Add a short section in README.md on how to test the tool manually via

MCP Inspector (npx @modelcontextprotocol/inspector) as well as via the
provided mcp_client.py.

Do not fabricate a database — assume it already exists with the schema above,

and just provide connection code. If you need me to run any SQL to create
sample tables/data for local testing, provide that as a separate
schema.sql / seed.sql file.

Thought for 39s (ctrl+o to expand)

Claude Model

Other

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.227

Platform

Other

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

_No response_