mirror of
https://github.com/hanzoai/engine.git
synced 2026-11-20 14:17:52 +00:00
Replace all upstream project references (EricLBuehler/mistral.rs) with hanzoai/engine across 99 files. Changes are metadata, docs, and branding only — no code changes. - Cargo.toml: homepage/repository URLs -> github.com/hanzoai/engine - Cargo.toml: add Hanzo AI as primary author, keep original for attribution - Dockerfiles: WORKDIR /mistralrs -> /engine - docs/book.toml: title, authors, URLs updated - docs/*.md: all GitHub links repointed, product name -> Hanzo Engine - crate README.md files: rebranded to Hanzo Engine - hanzo-engine-dso/Cargo.toml: candle deps -> hanzo-ml workspace paths - .github/workflows/release.yml: remove upstream reference in changelog - Remove broken upstream GitHub asset link from ANYMOE.md - Replace ericlbuehler.github.io docs URLs with docs.hanzo.ai - Replace upstream Docker image refs with ghcr.io/hanzoai/engine - Replace raw.githubusercontent.com URLs with hanzoai/engine/main - Replace master branch refs with main in docs links Preserved: - LICENSE file (legally required original copyright) - Cargo crate names (mistralrs-*) — renaming would break builds - Rust/Python module names in code examples - Environment variable names (MISTRALRS_*) from source code - pyproject.toml package name (tied to pyo3 module name)
3.7 KiB
3.7 KiB
MCP Transport Types
Hanzo Engine supports three transport types for connecting to MCP servers, each optimized for different use cases.
HTTP Transport
Best for public APIs, RESTful services, and servers behind load balancers.
Configuration
{
"source": {
"type": "Http",
"url": "https://api.example.com/mcp",
"timeout_secs": 30,
"headers": {
"X-API-Version": "v1",
"User-Agent": "hanzo-engine/0.7.0"
}
},
"bearer_token": "your-api-token"
}
Features
- Server-Sent Events (SSE) support for streaming responses
- Custom headers for API versioning or client identification
- Bearer token authentication (added as
Authorization: Bearer <token>) - Configurable timeouts
- Standard HTTP semantics
Example: Hugging Face MCP
McpServerSource::Http {
url: "https://hf.co/mcp".to_string(),
timeout_secs: Some(30),
headers: None,
}
WebSocket Transport
Best for real-time applications, bidirectional communication, and low-latency requirements.
Configuration
{
"source": {
"type": "WebSocket",
"url": "wss://realtime.example.com/mcp",
"timeout_secs": 60,
"headers": {
"Origin": "https://engine.hanzo.ai",
"Sec-WebSocket-Protocol": "mcp"
}
},
"bearer_token": "your-websocket-token"
}
Features
- Persistent connections reduce handshake overhead
- Server-initiated notifications
- Lower latency for frequent tool calls
- Automatic reconnection handling
- WebSocket-specific headers support
Example: Real-time Data Feed
McpServerSource::WebSocket {
url: "wss://data.example.com/mcp".to_string(),
timeout_secs: Some(60),
headers: Some(headers),
}
Process Transport
Best for local tools, development servers, and sandboxed environments.
Configuration
{
"source": {
"type": "Process",
"command": "mcp-server-filesystem",
"args": ["--root", "/tmp", "--readonly"],
"work_dir": "/home/user/workspace",
"env": {
"MCP_LOG_LEVEL": "info",
"MCP_TIMEOUT": "30"
}
}
}
Features
- No network overhead
- Process isolation for security
- Direct stdin/stdout communication
- Environment variable configuration
- Working directory control
- No authentication needed (process inherits permissions)
Example: Filesystem Server
McpServerSource::Process {
command: "mcp-server-filesystem".to_string(),
args: vec!["--root".to_string(), "/tmp".to_string()],
work_dir: None,
env: None,
}
Transport Selection Guide
| Use Case | Recommended Transport | Why |
|---|---|---|
| Public APIs | HTTP | Standard auth, caching, load balancing |
| Local tools | Process | No network, process isolation |
| Real-time data | WebSocket | Low latency, server push |
| Corporate proxies | HTTP | Proxy support, standard ports |
| Development | Process | Easy debugging, no network setup |
| Interactive apps | WebSocket | Bidirectional, persistent connection |
Security Considerations
HTTP
- Always use HTTPS in production
- Bearer tokens transmitted with each request
- Consider token rotation strategies
WebSocket
- Use WSS (WebSocket Secure) in production
- Bearer token sent during handshake
- Connection persists with authenticated state
Process
- Inherits user permissions
- Sandboxing via work_dir and env
- No network exposure
Performance Tips
- HTTP: Enable keep-alive, use connection pooling
- WebSocket: Reuse connections, handle reconnection gracefully
- Process: Minimize startup time, use long-running processes
Error Handling
All transports implement automatic retry with exponential backoff:
- Initial retry: 1 second
- Max retry: 60 seconds
- Max attempts: 5
Custom retry behavior can be configured per server.