@bulkhead-ai/server

@floatingsidewal/bulkhead-server

HTTP REST and MCP server for the Bulkhead content protection engine.

Install

Then install:

npm install @floatingsidewal/bulkhead-server

MCP Server

Exposes three tools via the Model Context Protocol (stdio transport):

Tool	Description
bulkhead_scan	Scan text for PII, secrets, injection. Modes: fast (regex), model (regex+BERT), deep (full cascade)
bulkhead_redact	Scan and return redacted text with [REDACTED-TYPE] placeholders
bulkhead_configure	Enable/disable guards at runtime

Claude Code

Add to your project's .mcp.json:

{
  "mcpServers": {
    "bulkhead": {
      "command": "bulkhead-mcp"
    }
  }
}

Or run directly:

npx bulkhead-mcp

GitHub Copilot

Add to .github/copilot/mcp.json with the same format.

HTTP REST Server

bulkhead-server
# or
npx bulkhead-server

Endpoints

Method	Path	Description
POST	/v1/scan	Regex-only scan (sub-ms)
POST	/v1/scan/model	Regex + BERT
POST	/v1/scan/deep	Full cascade (regex + BERT + LLM)
POST	/v1/redact	Scan and redact
GET	/healthz	Health check
GET	/readyz	Readiness (BERT model loaded?)
GET	/info	Guard configuration summary

Request Format

curl -X POST http://localhost:3000/v1/scan \
  -H "Content-Type: application/json" \
  -d '{"text": "My SSN is 123-45-6789"}'

Environment Variables

Variable	Default	Description
BULKHEAD_PORT	3000	Server port
BULKHEAD_HOST	0.0.0.0	Bind address
BULKHEAD_API_KEY	--	Enable API key auth (via X-API-Key header)
BULKHEAD_LOG_LEVEL	info	Log level
BULKHEAD_CASCADE_MODEL_ENABLED	false	Enable BERT layer
BULKHEAD_LLM_PROVIDER	--	LLM provider: openai, anthropic, or custom
BULKHEAD_LLM_API_KEY	--	LLM provider API key
BULKHEAD_LLM_ENDPOINT	--	Custom LLM endpoint URL

Docker

# HTTP mode
docker run -p 3000:3000 bulkhead

# MCP mode (stdio)
docker run -i bulkhead packages/server/dist/mcp/index.js

See the deployment guide for Docker configuration details.

License

MIT