This page collects setup notes for common MCP clients.

Before You Configure A Client

The project-operated public endpoint and website chatbot are retired. Use a local stdio client, or supply your own HTTP deployment. Remote-only clients require an operator-provided, reachable HTTPS URL.

Local Installation

Prefer npx in client configs when you need a local server. It avoids hard-coding an install path and works across macOS, Linux, and Windows.

Run locally without a global install:

npx -y fdic-mcp-server

If you prefer a global install:

npm install -g fdic-mcp-server

Client support changes quickly. Treat the linked official docs as the source of truth.

Last verified: March 15, 2026.

Claude Code

Claude Code has a plugin system that installs both the MCP server and Claude Code skills in one step.

Add the marketplace and install:

/plugin marketplace add jflamb/fdic-mcp-server
/plugin install fdic-mcp-server@fdic-mcp-server

This gives you:

Skills complement the MCP tools. Tools give you raw data access; skills build structured, multi-step workflows on top of those tools. See Skills for details.

Manual local setup without the plugin:

claude mcp add fdic -- npx -y fdic-mcp-server

For a self-hosted HTTP server (replace the example URL with your own):

claude mcp add fdic --transport http https://your-server.example/mcp

Or add directly to your project or user .mcp.json:

{
  "mcpServers": {
    "fdic": {
      "url": "https://your-server.example/mcp"
    }
  }
}

Claude Desktop

Use local stdio with Node.js 20 or later. If you operate an HTTPS MCP endpoint, you can instead configure your own remote connector. The project provides no hosted URL.

Config file:

~/Library/Application Support/Claude/claude_desktop_config.json

Example:

{
  "mcpServers": {
    "fdic": {
      "command": "npx",
      "args": ["-y", "fdic-mcp-server"]
    }
  }
}

After editing the file, restart Claude Desktop.

Official docs:

ChatGPT

ChatGPT Developer Mode supports MCP apps/connectors, but it expects a remote MCP server over streaming HTTP or SSE, not a local stdio process.

Use your own HTTPS MCP URL (example placeholder):

https://your-server.example/mcp

In ChatGPT:

  1. Go to Settings -> Apps -> Advanced settings -> Developer mode and enable Developer mode.
  2. Open the Apps settings page and create an app for your MCP server.
  3. Use https://your-server.example/mcp.
  4. Refresh tools from the app details page if needed.

Local ChatGPT App Validation

Use this loop when changing tool descriptors, app resources, widget HTML, or ChatGPT-specific metadata.

Build and start the HTTP server:

npm install
npm run build
TRANSPORT=http PORT=3000 node dist/index.js

Inspect the local endpoint with MCP Inspector before opening ChatGPT:

http://127.0.0.1:3000/mcp

Then expose the local server through HTTPS:

ngrok http 3000

Create or update the ChatGPT Developer Mode app with:

https://<your-ngrok-host>/mcp

After any change to tool descriptors, input schemas, app resources, CSP metadata, or widget URI versions, refresh the app/tools from the ChatGPT app details page. ChatGPT may cache widget resources by URI, so use a new ui://widget/...-vN.html URI for breaking widget changes.

Suggested smoke prompts:

  • Search for Bank of America and cite the result.
  • Fetch institution:3511.
  • Show a deep dive for CERT 3511.
  • Show branches for CERT 3511.
  • What data date basis are you using?

If you independently operate and submit a ChatGPT app, provide your own hosting and app materials:

  • stable public HTTPS /mcp endpoint
  • app name, description, icon, and screenshots
  • privacy policy and support URLs
  • test prompts and expected outputs for review
  • clear disclosure that public health scoring is not an official CAMELS rating
  • production logs or alerts for tool errors and latency

Notes:

  • ChatGPT supports streaming HTTP and SSE for MCP apps.
  • ChatGPT Developer Mode availability depends on plan and workspace settings.
  • For Business and Enterprise/Edu workspaces, admins may need to allow custom apps first.
  • The search and fetch tools are always on for ChatGPT-compatible retrieval, including institution, failure, branch, and schema results.

Official docs:

Gemini CLI

Gemini CLI supports MCP servers through ~/.gemini/settings.json for user scope or .gemini/settings.json for project scope.

Self-hosted endpoint example (replace the placeholder):

{
  "mcpServers": {
    "fdic": {
      "httpUrl": "https://your-server.example/mcp"
    }
  }
}

Use this HTTP configuration only when you operate the endpoint. For local setup, use stdio below.

Local stdio:

Config example:

{
  "mcpServers": {
    "fdic": {
      "command": "npx",
      "args": ["-y", "fdic-mcp-server"],
      "trust": true
    }
  }
}

Verification:

gemini mcp list

Notes:

  • Gemini CLI supports both local stdio and remote HTTP MCP servers.
  • If the current folder is untrusted, local stdio MCP servers may appear disconnected until you trust the folder.

Official docs:

GitHub Copilot CLI

GitHub Copilot CLI supports local MCP servers through ~/.copilot/mcp-config.json.

Config example:

{
  "mcpServers": {
    "fdic": {
      "type": "local",
      "command": "npx",
      "args": ["-y", "fdic-mcp-server"],
      "env": {},
      "tools": ["*"]
    }
  }
}

CLI-based setup:

/mcp add

Or edit the config file directly, then check:

/mcp show

Notes:

  • The built-in GitHub MCP server is separate; these instructions are only for adding this FDIC server.
  • Copilot CLI makes newly added MCP servers available immediately without a restart.

Official docs:

Other MCP Hosts

Many other MCP hosts accept the same basic stdio pattern:

{
  "mcpServers": {
    "fdic": {
      "command": "npx",
      "args": ["-y", "fdic-mcp-server"]
    }
  }
}

Before documenting a specific host in this repo, verify:

  • that it currently supports MCP
  • whether it expects local stdio, remote HTTP, or both
  • where its config file lives
  • whether it requires a restart or explicit enable step