The Developer's Dilemma
You're debugging a production issue. The stack trace includes customer email addresses. The fastest solution? Paste it into Claude and ask for help.
But that customer data is now:
- Stored in Anthropic's systems
- Potentially used for training (depending on your plan)
- Visible to anyone with access to your chat history
77% of developers paste sensitive data into AI tools. Most don't realize the implications until it's too late. Beyond individual mistakes, 34.8% of all ChatGPT inputs in enterprise environments contain confidential business information — and organisational policies alone cannot stop it: 77% of employees share sensitive data with AI tools even when prohibitive policies are in place.
Why policies fail
A government contractor uploaded FEMA applicant personal and health data to ChatGPT for processing — despite a clear policy against it. Samsung suffered three source code leaks within one month after lifting a ChatGPT ban. Human error is the #1 cause of AI-related data leaks. Technical controls that operate automatically are the only reliable defence.
Source: anonym.community — From FEMA to Finance (March 2026)
The MCP Server Solution
The Model Context Protocol (MCP) Server acts as a transparent proxy between you and AI assistants. It:
- Intercepts your prompts before they reach the AI
- Detects and anonymizes PII automatically
- Sends only anonymized data to the AI
- De-anonymizes responses so you see original context
You interact with AI normally — the protection is invisible.
Setting Up MCP Server
Prerequisites
- Claude Desktop, Cursor, or VS Code with Claude extension
- anonym.legal account on the Pro or Business plan (MCP Server requires a paid plan)
- anonym.legal API key — get one at anonym.legal
No Node.js or npm install needed — anonym.legal's MCP Server is HTTP transport only.
Step 1: Get Your API Key
- Sign up at anonym.legal/auth/signup
- Navigate to Settings → API Tokens
- Generate a new token
- Copy the token (you'll only see it once)
Step 2: Configure Claude Desktop
Edit your Claude Desktop configuration file:
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
Linux: ~/.config/Claude/claude_desktop_config.json
{
"mcpServers": {
"anonym-legal": {
"type": "http",
"url": "https://anonym.legal/mcp",
"headers": {
"Authorization": "Bearer your-api-key-here"
}
}
}
}
Step 3: Restart Claude Desktop
Close and reopen Claude Desktop. You should see "anonym-legal" in the MCP servers list.
Configuration for Cursor IDE
Cursor uses the same MCP protocol. Add to your Cursor settings:
{
"mcp": {
"servers": {
"anonym-legal": {
"url": "https://anonym.legal/mcp",
"headers": {
"Authorization": "Bearer your-api-key-here"
}
}
}
}
}
What Gets Anonymized
The MCP Server detects and anonymizes 267+ entity types:
| Category | Examples |
|---|---|
| Personal | Names, emails, phone numbers, DOB |
| Financial | Credit cards, bank accounts, IBANs |
| Government | SSNs, passport numbers, driver's licenses |
| Technical | IP addresses, API keys, tokens |
| Healthcare | MRNs, patient IDs, insurance numbers |
| Corporate | Employee IDs, account numbers |
Example Transformation
Your prompt:
Debug this error from user john.smith@acme.com:
Error: Payment failed for card 4532-1234-5678-9012
Customer ID: CUST-12345
IP: 192.168.1.100
What Claude sees:
Debug this error from user [EMAIL_1]:
Error: Payment failed for card [CREDIT_CARD_1]
Customer ID: [CUSTOMER_ID_1]
IP: [IP_ADDRESS_1]
Claude's response (to you):
The error for john.smith@acme.com suggests the card
4532-1234-5678-9012 may have insufficient funds...
You see the original data; Claude only ever saw tokens.
Advanced Configuration
Custom Entity Types
Define additional patterns to detect:
{
"env": {
"ANONYM_API_KEY": "your-api-key",
"CUSTOM_PATTERNS": "JIRA-[0-9]+,TICKET-[A-Z0-9]+"
}
}
Allowlist Specific Data
Some data shouldn't be anonymized (public company names, product names):
{
"env": {
"ANONYM_API_KEY": "your-api-key",
"ALLOWLIST": "Anthropic,Claude,anonym.legal"
}
}
Security Considerations
| Component | Location |
|---|---|
| MCP Server | Your machine (local) |
| PII Detection API | anonym.legal / cloak.business servers (Germany, Hetzner) |
| AI Model | Anthropic/OpenAI servers |
anonym.legal does not store your prompts or responses. The API receives text for analysis, returns entity positions, and immediately discards the input. See the privacy policy for details.
Pricing
anonym.legal plans — MCP Server requires Pro or Business:
| Plan | Tokens/month | Price |
|---|---|---|
| Free | 200 | €0 |
| Basic | 1,000 | €3/month |
| Pro | 4,000 | €15/month |
| Business | 10,000 | €29/month |
The MCP Server is available on Pro (€15/month) and Business (€29/month) plans.
Conclusion
AI assistants are essential for modern development, but they don't need to see your customers' data.
The MCP Server integration requires no workflow changes, protects PII automatically, and works with Claude Desktop, Cursor, and VS Code. Set it up once — your data is protected by default.
Limitations
The MCP Server itself runs locally, but PII detection is not fully offline — and access has real per-plan constraints.
- PII detection calls the anonym.legal API in Germany — this adds a network dependency and a small per-call latency that a fully offline setup would not have.
- Free (200 tokens) and Basic (1,000 tokens) plans do not include MCP access; Pro or Business is required, a real cost for individual developers evaluating the workflow.
- Setup is per MCP client (Claude Desktop, Cursor, VS Code) — each tool and each contributor needs its own configuration and API key; it isn't a single organisation-wide switch.
Sources
- MCP Protocol Specification — modelcontextprotocol.io
- LayerX — 77% Developer AI Usage
- anonym.community — From FEMA to Finance: Why AI Policy Without Technical Controls Fails (March 2026) — 34.8% confidential ChatGPT inputs, FEMA contractor case
- anonym.community — Beyond the ChatGPT Ban (March 2026) — Samsung three-leak case study