Quick Checks

Before debugging a client-specific issue, verify:

• Node.js is version 20 or later
• the package is installed or the local build exists
• the MCP host supports the transport you configured
• your client config points at the correct binary or HTTP URL
• your prompt matches the FDIC dataset and time basis you actually need

FAQ

Why does ChatGPT not connect to my local binary?

ChatGPT Developer Mode expects a remotely reachable MCP server over HTTP or SSE. A local stdio command such as /opt/homebrew/bin/fdic-mcp-server is not enough by itself.

Use the hosted URL when possible:

https://bankfind.jflamb.com/mcp

If you are intentionally running your own local or self-hosted HTTP version, start it with:

TRANSPORT=http PORT=3000 fdic-mcp-server

Then expose it through a reachable HTTPS URL.

If you are running the Docker image or Cloud Run deployment instead of the local binary, use port 8080 rather than 3000.

Why do I get no results for a bank I know exists?

Common causes:

• the filter or sort choice does not apply to the kind of records you asked for
• the bank is inactive and you filtered on ACTIVE:1
• the report date is outside the relevant range
• you asked a branch question with a quarterly dataset instead of SOD or locations data

Why do results look inconsistent across tools?

The datasets answer different questions and do not all update on the same schedule.

• financials and demographics are quarterly
• summary data is annual
• SOD is annual branch-level data as of June 30
• failures and history use their own event dates

If you compare across datasets, state the date basis clearly.

Why does a comparison return warnings?

Warnings usually mean the comparison has limits you should read before drawing conclusions, such as partial peer coverage or missing values for some institutions.

The tool still returns data, but you should read the warnings before drawing conclusions.

Why does peer group ranking denominator change by metric?

Rankings are metric-specific. If some peers do not have a valid value for ROA, efficiency ratio, or another metric, they are excluded from that metric’s denominator.

Troubleshooting By Symptom

The MCP host says the server failed to start

Check:

• whether the command path exists
• whether the package was installed under a different npm prefix
• whether you built the project before pointing a client at dist/index.js

Useful commands:

which fdic-mcp-server
node dist/index.js

The HTTP client connects but tool calls fail

Check:

• that the server is running with TRANSPORT=http
• that your client is pointed at the MCP URL rather than the site home page
• that the URL ends in /mcp
• that your MCP host supports remote HTTP MCP connections

The model keeps choosing the wrong tool

Improve the prompt by stating:

• whether you want raw records or analysis
• whether you already know the CERT
• the date basis
• the ranking metric or geography

See Prompting Guide and Tool Reference.

I asked about branches and got quarterly figures instead

You likely needed branch or office data instead of quarterly financials.

• Use fdic_search_sod for branch deposit totals
• Use fdic_search_demographics for office counts and market-structure fields
• Use fdic_search_locations for branch records

A local client cannot find the binary after global install

Your npm global prefix may differ from the examples. Run:

npm prefix -g
which fdic-mcp-server

Then update the client configuration with the actual binary path.

Still Stuck

• Review Client Setup
• Review Support
• Open an issue with the exact MCP host, setup path, prompt, and observed error