CLASP - Claude Language Agent Super Proxy

A high-performance Go proxy that translates Claude/Anthropic API calls to OpenAI-compatible endpoints, enabling Claude Code to work with any LLM provider.

Features

• Bundled Claude Code: Automatically includes Claude Code as a dependency - single npx clasp-ai installs everything
• Multi-Provider Support: OpenAI, Azure OpenAI, OpenRouter (200+ models), and custom endpoints (Ollama, vLLM, LM Studio)
• Full Protocol Translation: Anthropic Messages API ↔ OpenAI Chat Completions API
• SSE Streaming: Real-time token streaming with state machine processing
• Tool Calls: Complete translation of tool_use/tool_result between formats
• Connection Pooling: Optimized HTTP transport with persistent connections
• Retry Logic: Exponential backoff for transient failures
• Metrics Endpoint: Request statistics and performance monitoring
• API Key Authentication: Secure the proxy with optional API key validation

Installation

Via npm (recommended)

# Install globally
npm install -g clasp-ai

# Or run directly with npx
npx clasp-ai

Via Go

go install github.com/jedarden/clasp/cmd/clasp@latest

From Source

git clone https://github.com/jedarden/CLASP.git
cd CLASP
make build

Via Docker

# Run with Docker (from GitHub Container Registry)
docker run -d -p 8080:8080 \
  -e OPENAI_API_KEY=sk-... \
  ghcr.io/jedarden/clasp:latest

# With specific version
docker run -d -p 8080:8080 \
  -e OPENAI_API_KEY=sk-... \
  ghcr.io/jedarden/clasp:0.24.8

# Or with docker-compose
docker-compose up -d

Available Docker tags:

• latest - Latest stable release
• 0.24 - Latest 0.24.x release
• 0.24.8 - Specific version

Quick Start

Using with OpenAI

# Set your API key
export OPENAI_API_KEY=sk-...

# Start the proxy
clasp -model gpt-4o

# In another terminal, use Claude Code
ANTHROPIC_BASE_URL=http://localhost:8080 claude

Using with Azure OpenAI

export AZURE_API_KEY=your-key
export AZURE_OPENAI_ENDPOINT=https://your-resource.openai.azure.com
export AZURE_DEPLOYMENT_NAME=gpt-4

clasp -provider azure

Using with OpenRouter

export OPENROUTER_API_KEY=sk-or-...

clasp -provider openrouter -model anthropic/claude-3-sonnet

Using with Local Models (Ollama)

export CUSTOM_BASE_URL=http://localhost:11434/v1

clasp -provider custom -model llama3.1

Configuration

CLASP supports three configuration methods, with the following precedence (highest to lowest):

1. Command-line flags - Override all other settings
2. Environment variables - Override config file settings
3. Configuration file - YAML-based configuration
4. Default values - Built-in defaults

Configuration File

CLASP can load configuration from a YAML file, which is cleaner for complex setups and version control friendly.

Configuration file locations (searched in order):

1. ./clasp.yaml or ./clasp.yml - Current directory
2. ~/.clasp/config.yaml - User config directory
3. ~/.config/clasp/config.yaml - XDG config directory
4. /etc/clasp/config.yaml - System-wide config
5. Custom path via CLASP_CONFIG_FILE environment variable

Example configuration file (clasp.yaml):

# Provider selection
provider: openai

# API Keys (supports ${VAR} syntax for environment variables)
api_keys:
  openai: ${OPENAI_API_KEY}
  # azure: ${AZURE_API_KEY}
  # openrouter: ${OPENROUTER_API_KEY}

# Custom endpoints (optional)
endpoints:
  openai: https://api.openai.com/v1
  ollama: http://localhost:11434

# Model configuration
models:
  default: gpt-4o
  opus: gpt-4o           # Maps claude-opus-* to gpt-4o
  sonnet: gpt-4o-mini    # Maps claude-sonnet-* to gpt-4o-mini
  haiku: gpt-4o-mini     # Maps claude-haiku-* to gpt-4o-mini

# Server settings
server:
  port: 8080
  log_level: info

# Feature toggles
cache:
  enabled: true
  max_size: 1000
  ttl: 3600

rate_limit:
  enabled: false
  requests: 60
  window: 60

# Multi-provider routing
multi_provider:
  enabled: false
  opus:
    provider: openai
    model: gpt-4o
  sonnet:
    provider: anthropic
    model: claude-sonnet-4-20250514

Environment variable expansion:

The config file supports ${VAR} and ${VAR:-default} syntax:

api_keys:
  openai: ${OPENAI_API_KEY}
  openrouter: ${OPENROUTER_API_KEY:-sk-or-default-key}

Complete example: See clasp.example.yaml for all available options.

Command Line Options

clasp [options]

Options:
  -port <port>           Port to listen on (default: 8080)
  -provider <name>       LLM provider: openai, azure, openrouter, custom
  -model <model>         Default model to use for all requests
  -debug                 Enable debug logging (full request/response)
  -rate-limit            Enable rate limiting
  -cache                 Enable response caching
  -cache-max-size <n>    Maximum cache entries (default: 1000)
  -cache-ttl <n>         Cache TTL in seconds (default: 3600)
  -multi-provider        Enable multi-provider tier routing
  -fallback              Enable fallback routing for auto-failover
  -auth                  Enable API key authentication
  -auth-api-key <key>    API key for authentication (required with -auth)
  -version               Show version information
  -help                  Show help message

Environment Variables

Model Mapping

CLASP can automatically map Claude model tiers to your provider's models:

# Map Claude tiers to specific models
export CLASP_MODEL_OPUS=gpt-4o
export CLASP_MODEL_SONNET=gpt-4o-mini
export CLASP_MODEL_HAIKU=gpt-3.5-turbo

Multi-Provider Routing

Route different Claude model tiers to different LLM providers for cost optimization:

# Enable multi-provider routing
export CLASP_MULTI_PROVIDER=true

# Route Opus tier to OpenAI (premium)
export CLASP_OPUS_PROVIDER=openai
export CLASP_OPUS_MODEL=gpt-4o
export CLASP_OPUS_API_KEY=sk-...  # Optional, inherits from OPENAI_API_KEY

# Route Sonnet tier to OpenRouter (cost-effective)
export CLASP_SONNET_PROVIDER=openrouter
export CLASP_SONNET_MODEL=anthropic/claude-3-sonnet
export CLASP_SONNET_API_KEY=sk-or-...  # Optional, inherits from OPENROUTER_API_KEY

# Route Haiku tier to local Ollama (free)
export CLASP_HAIKU_PROVIDER=custom
export CLASP_HAIKU_MODEL=llama3.1
export CLASP_HAIKU_BASE_URL=http://localhost:11434/v1

clasp -multi-provider

Multi-Provider Environment Variables:

Where {TIER} is OPUS, SONNET, or HAIKU.

Benefits:

• Cost Optimization: Use expensive providers only for complex tasks
• Latency Reduction: Route simple requests to faster local models
• Redundancy: Mix cloud and local providers for reliability
• A/B Testing: Compare different models across tiers

API Endpoints

Supported Tools

CLASP supports all Claude Code 2.1.34+ tools with full parameter validation and OpenAI-compatible schema transformation. See docs/api-reference/claude-code-tools.md for the complete tool reference including:

• File Operations: Read, Write, Edit, Glob, Grep
• Command Execution: Bash with background support
• Web Operations: WebSearch, WebFetch
• Agent Orchestration: Task, TaskOutput, TaskStop (with model/resume/max_turns parameters)
• Interactive Features: AskUserQuestion, ExitPlanMode
• Specialized Tools: NotebookEdit (Jupyter), LSP (code intelligence), Skill invocation
• Task Management: TaskCreate, TaskGet, TaskUpdate, TaskList (CLI-only)

Example Usage

With curl

curl http://localhost:8080/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: any-key" \
  -d '{
    "model": "claude-3-5-sonnet-20241022",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Hello!"}
    ]
  }'

Streaming

curl http://localhost:8080/v1/messages \
  -H "Content-Type: application/json" \
  -H "x-api-key: any-key" \
  -d '{
    "model": "claude-3-5-sonnet-20241022",
    "max_tokens": 1024,
    "stream": true,
    "messages": [
      {"role": "user", "content": "Count to 5"}
    ]
  }'

Response Caching

CLASP can cache responses to reduce API costs and improve latency for repeated requests:

# Enable caching with defaults (1000 entries, 1 hour TTL)
clasp -cache

# Custom cache settings
clasp -cache -cache-max-size 500 -cache-ttl 1800

# Via environment
CLASP_CACHE=true CLASP_CACHE_MAX_SIZE=500 clasp

Caching behavior:

• Only non-streaming requests are cached
• Requests with temperature > 0 are not cached (non-deterministic)
• Cache uses LRU (Least Recently Used) eviction when full
• Cache entries expire after TTL (time-to-live)
• Response headers include X-CLASP-Cache: HIT or X-CLASP-Cache: MISS

Metrics

Access /metrics for request statistics:

{
  "requests": {
    "total": 100,
    "successful": 98,
    "errors": 2,
    "streaming": 75,
    "tool_calls": 15,
    "success_rate": "98.00%"
  },
  "performance": {
    "avg_latency_ms": "523.50",
    "requests_per_sec": "2.34"
  },
  "cache": {
    "enabled": true,
    "size": 42,
    "max_size": 1000,
    "hits": 156,
    "misses": 44,
    "hit_rate": "78.00%"
  },
  "uptime": "5m30s"
}

Docker

Build and Run

# Build Docker image
make docker

# Run container
make docker-run

# Stop container
make docker-stop

Docker Compose

Create a .env file with your configuration:

PROVIDER=openai
OPENAI_API_KEY=sk-...
CLASP_DEFAULT_MODEL=gpt-4o

Then start the service:

docker-compose up -d

Docker Environment Variables

All configuration is done through environment variables. See the Environment Variables section above.

Development

# Build
make build

# Run tests
make test

# Build for all platforms
make build-all

# Build Docker image
make docker

# Format code
make fmt

Debugging

Enable debug logging to troubleshoot issues:

# Via CLI flag
clasp -debug

# Via environment variable
CLASP_DEBUG=true clasp

# Log only requests or responses
CLASP_DEBUG_REQUESTS=true clasp
CLASP_DEBUG_RESPONSES=true clasp

Debug output includes:

• Incoming Anthropic requests
• Outgoing OpenAI requests
• Raw OpenAI responses
• Transformed Anthropic responses

Authentication

Secure your CLASP proxy with API key authentication to control access:

# Enable authentication with CLI flags
clasp -auth -auth-api-key "my-secret-key"

# Or via environment variables
CLASP_AUTH=true CLASP_AUTH_API_KEY="my-secret-key" clasp

Providing the API Key

Clients can provide the API key in two ways:

# Via x-api-key header
curl http://localhost:8080/v1/messages \
  -H "x-api-key: my-secret-key" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-3-5-sonnet-20241022", ...}'

# Via Authorization header (Bearer token)
curl http://localhost:8080/v1/messages \
  -H "Authorization: Bearer my-secret-key" \
  -H "Content-Type: application/json" \
  -d '{"model": "claude-3-5-sonnet-20241022", ...}'

Authentication Options

Endpoint Access with Authentication Enabled

Using with Claude Code

When authentication is enabled, set both the base URL and API key:

# Start CLASP with auth
OPENAI_API_KEY=sk-... clasp -auth -auth-api-key "proxy-key"

# Use with Claude Code (the proxy key is passed as the Anthropic key)
ANTHROPIC_BASE_URL=http://localhost:8080 ANTHROPIC_API_KEY=proxy-key claude

Request Queuing

Queue requests during provider outages for automatic retry:

# Enable request queuing
clasp -queue

# Custom queue settings
clasp -queue -queue-max-size 200 -queue-max-wait 60

# Via environment
CLASP_QUEUE=true CLASP_QUEUE_MAX_SIZE=200 clasp

Queue Options

Circuit Breaker

Prevent cascade failures with circuit breaker pattern:

# Enable circuit breaker
clasp -circuit-breaker

# Custom circuit breaker settings
clasp -circuit-breaker -cb-threshold 10 -cb-recovery 3 -cb-timeout 60

# Via environment
CLASP_CIRCUIT_BREAKER=true clasp

Circuit Breaker Options

Circuit Breaker States

• Closed: Normal operation, requests pass through
• Open: Circuit tripped, requests fail fast with 503
• Half-Open: Testing if service recovered, limited requests allowed

Maximum Resilience Configuration

For production deployments requiring maximum resilience:

# Enable queue + circuit breaker + fallback
OPENAI_API_KEY=sk-xxx OPENROUTER_API_KEY=sk-or-xxx \
  clasp -queue -circuit-breaker -fallback \
    -queue-max-size 200 \
    -cb-threshold 5 \
    -cb-timeout 30

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please open an issue or submit a pull request on GitHub.