Documentation

Integrations

Pretense is a drop-in proxy. Connect it to any AI coding tool by changing one setting.

Claude Code

Recommended

Route every Claude Code session through Pretense by setting a single environment variable. All prompts are mutated before reaching the Anthropic API.

  1. 1Install and start the Pretense proxy (pretense start)
  2. 2Set ANTHROPIC_BASE_URL in your shell profile or session
  3. 3Run Claude Code as normal -- mutation happens transparently
Shell
# Option 1: per-session
ANTHROPIC_BASE_URL=http://localhost:9339 claude "refactor auth.ts"

# Option 2: export in ~/.zshrc or ~/.bashrc
export ANTHROPIC_BASE_URL=http://localhost:9339

Cursor

Override the model base URL in Cursor settings to route requests through Pretense. Works with GPT-4, Claude, and any OpenAI-compatible model.

  1. 1Open Cursor and go to Settings (Cmd+,)
  2. 2Navigate to Models and find "Override OpenAI Base URL"
  3. 3Set it to http://localhost:9339
  4. 4Your existing API key works unchanged
Shell
# Alternatively, set the env var before launching Cursor
OPENAI_BASE_URL=http://localhost:9339 cursor

VS Code Extension

The Pretense VS Code extension adds a status bar indicator, a scan command, and a mutation map viewer. Install from the marketplace.

  1. 1Search for "Pretense" in the VS Code Extensions panel
  2. 2Install and reload VS Code
  3. 3The status bar shows live mutation count and proxy status
  4. 4Use Cmd+Shift+P and type "Pretense: Scan" to scan the active file

GitHub Copilot

Enterprise

Enterprise Copilot deployments can be routed through Pretense via a proxy configuration in your organization settings. Contact sales for setup assistance.

  1. 1Requires GitHub Copilot for Business or Enterprise
  2. 2Configure your organization-level proxy endpoint to http://your-pretense-server:9339
  3. 3All Copilot requests flow through Pretense before reaching GitHub
  4. 4Contact hello@pretense.ai for enterprise proxy setup documentation

MCP Server

Claude Code

The Pretense MCP server registers with Claude Code as a tool. Every prompt is automatically scanned before being sent, with no changes to your workflow.

  1. 1Install Pretense globally (npm install -g pretense)
  2. 2Add the MCP server configuration to .claude/settings.json
  3. 3Reload Claude Code -- the pretense_scan tool is now available
  4. 4Claude Code automatically calls pretense_scan before sending prompts
JSON
// .claude/settings.json
{
  "mcpServers": {
    "pretense": {
      "command": "pretense-mcp",
      "args": []
    }
  }
}

CI/CD Pipeline

Block any unprotected LLM API call from reaching production. The Pretense CLI integrates with any CI system that runs Node.js.

  1. 1Add pretense to your project devDependencies
  2. 2Run pretense scan ./src as a pre-deploy check
  3. 3Fail the build if the exit code is non-zero (secrets detected)
  4. 4Export the scan report as a CI artifact for auditing
JSON
# package.json scripts
{
  "scripts": {
    "scan": "pretense scan ./src --format json --output scan-report.json",
    "preflight": "npm run scan && echo 'Scan passed'"
  }
}

Need a different integration?

If your tool is not listed, email us and we will add official support. Pretense speaks OpenAI-compatible API format so most tools work already.

Was this page helpful?
Ask me anything