The ultimate Model Context Protocol integration platform
Zero-configuration AI agent infrastructure with enterprise-grade security
curl -fsSL https://get.cyrup.ai/install | shπ Installation Details
Cross-Platform Support: Auto-detects OS and architecture
- macOS: Intel + Apple Silicon (Darwin)
- Linux: x86_64 + aarch64 (systemd)
- Windows: x86_64 + i686 (services)
What it installs:
- Rust toolchain (if missing)
- Clones
git@github.com:cyrup-ai/sweetmcp - Builds native binaries with Cargo
- Generates wildcard SSL certificates (
*.cyrup.{dev,ai,cloud,pro}) - Configures OS trust stores and DNS resolution
- Deploys systemd/launchd services
- Starts daemon with proper privileges
Result: Production-ready MCP infrastructure in ~60 seconds
SweetMCP provides six foundational AI infrastructure components:
Context-augmentation memory system (vector & graph relationships)
- Backend: SurrealDB multi-model database
- Vector Search: Embedded similarity matching
- Graph Relations: Knowledge graph with entity linking
- Retention: Global + project-local scoping
- Integration: Full Obsidian knowledge-base support
- Visibility: Transparent agent context inspection
// Example: Memory API sweetmcp::memory::store_context("project_id", &embedding, &metadata).await?; let related = sweetmcp::memory::semantic_search("query", 0.8, 10).await?;
Post-quantum encryption vault and secret management
- Cryptography: Military-grade, quantum-resistant algorithms
- Zero-Knowledge: Agents never access raw credentials
- Architecture: Pure Rust implementation with timing-safe operations
- Access Control: Time-based expiration and role-based permissions
- Storage: Encrypted vault with secure key derivation
// Example: Secret access without exposure let result = cryypt::execute_with_secret("aws_key", |secret| { aws_client.authenticate(secret).call_api() }).await?; // Secret never leaves secure context
Pure-Rust AI voice system with predictive capabilities
- STT: faster-whisper speech-to-text
- Wake Words: Native wake word detection
- Prediction: Anticipatory text completion during speech
- TTS: Pure-Rust Candle port of Dia Voice
- Performance: Low-latency, on-device processing
- Quality: Expressive voice synthesis with emotional range
// Example: Voice interaction let transcription = fluent_voice::listen_with_prediction().await?; let response = fluent_voice::synthesize(&text, Voice::Expressive).await?;
Autonomous "ambient agents" for background optimization
- Architecture: Self-configuring agent swarm
- Autonomy: Zero direct user interaction required
- Capabilities:
- Prompt optimization and auto-tuning
- Task decomposition and parallelization
- Documentation procurement and indexing
- Preference learning and memorization
- Integration: Assists primary AI models transparently
// Example: Ambient agent coordination ambient::spawn_optimizer_for(primary_model_id).await?; ambient::background_task_decomposer::register(complex_task).await?;
Advanced reasoning and autocoding capabilities
- Reasoning: Deep reasoning chains with iterative refinement
- Search: MCTS (Monte Carlo Tree Search) for solution exploration
- Research: Deep research agents with multi-source synthesis
- Context: Entire codebase understanding and analysis
- Generation: Specialized models for different programming domains
// Example: Code generation with reasoning let solution = code_gen::reason_and_generate(CodeRequest { context: codebase_context, requirements: user_spec, search_strategy: SearchStrategy::MCTS, reasoning_depth: 5 }).await?;
Universal tool integration without configuration
Supported Platforms:
- Claude Desktop | Claude Code
- Windsurf | Cursor.AI
- VSCode | Zed
- Raycast | Cline | Roo Code
- + thousands more via MCP standard
Operating Systems: Linux, macOS, Windows (all architectures) Setup Time: Literally zero - auto-discovery and configuration
// Auto-generated MCP configuration { "mcpServers": { "sweetmcp": { "command": "sweetmcp", "args": ["--stdio"], "env": {} } } }
- MCP (Model Context Protocol) - Native first-class support
- JSON-RPC 2.0 - Full specification compliance
- GraphQL - Query-based tool invocation
- Cap'n Proto - High-performance binary protocol
- Latency: Sub-millisecond tool routing
- Throughput: 10K+ requests/second per core
- Memory: Minimal footprint with smart caching
- Scaling: Horizontal with service mesh discovery
- Authentication: JWT with configurable signing algorithms
- Transport: TLS 1.3 with optional mTLS
- Rate Limiting: Configurable per-endpoint throttling
- Isolation: Process-level separation for tool execution
https://sweetmcp.cyrup.dev:8443 # Primary development endpoint
https://sweetmcp.cyrup.ai:8443 # AI-optimized routing
https://sweetmcp.cyrup.cloud:8443 # Cloud services integration
https://sweetmcp.cyrup.pro:8443 # Professional tooling
# Core Configuration export SWEETMCP_JWT_SECRET="$(openssl rand -base64 32)" export SWEETMCP_TCP_BIND="0.0.0.0:8443" export SWEETMCP_METRICS_BIND="127.0.0.1:9090" # Performance Tuning export SWEETMCP_INFLIGHT_MAX=1000 export SWEETMCP_WORKER_THREADS=8 export SWEETMCP_CONNECTION_POOL_SIZE=100 # Discovery & Clustering export SWEETMCP_DNS_SERVICE="_sweetmcp._tcp.example.com" export SWEETMCP_DISCOVERY_TOKEN="cluster-shared-secret" export SWEETMCP_NODE_ID="$(hostname)"
_sweetmcp._tcp.example.com. 300 IN SRV 10 50 8443 node1.example.com.
_sweetmcp._tcp.example.com. 300 IN SRV 10 30 8443 node2.example.com.
_sweetmcp._tcp.example.com. 300 IN SRV 20 20 8443 node3.example.com.
# Linux (systemd) sudo systemctl start|stop|restart cyrupd sudo systemctl enable cyrupd # Auto-start on boot journalctl -u cyrupd -f # Live logs # macOS (launchd) sudo launchctl load|unload /Library/LaunchDaemons/com.cyrup.cyrupd.plist tail -f /var/log/cyrupd.log # Windows (sc) sc start|stop cyrupd sc config cyrupd start=auto
curl -X POST https://sweetmcp.cyrup.dev:8443/ \ -H "Authorization: Bearer $JWT_TOKEN" \ -H "Content-Type: application/json" \ -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "file_operations/read", "arguments": {"path": "/etc/hosts"} } }'
curl -X POST https://sweetmcp.cyrup.dev:8443/graphql \ -H "Authorization: Bearer $JWT_TOKEN" \ -H "Content-Type: application/graphql" \ -d 'query { tools { list { name description inputSchema } } }'
# Store context curl -X POST https://sweetmcp.cyrup.dev:8443/ \ -H "Authorization: Bearer $JWT_TOKEN" \ -d '{ "jsonrpc": "2.0", "method": "memory/store", "params": { "content": "User prefers TypeScript over JavaScript", "metadata": {"type": "preference", "user_id": "123"} } }' # Semantic search curl -X POST https://sweetmcp.cyrup.dev:8443/ \ -H "Authorization: Bearer $JWT_TOKEN" \ -d '{ "jsonrpc": "2.0", "method": "memory/search", "params": { "query": "programming language preferences", "threshold": 0.8, "limit": 5 } }'
git clone git@github.com:cyrup-ai/sweetmcp.git cd sweetmcp cargo build --release cargo test --all-features
RUST_LOG=debug,sweetmcp=trace cargo run -- daemon --debug
# Service health curl -f https://sweetmcp.cyrup.dev:8443/health # Metrics (Prometheus format) curl http://127.0.0.1:9090/metrics # Tool registry status curl https://sweetmcp.cyrup.dev:8443/tools/list
# Stop services sudo systemctl stop cyrupd && sudo systemctl disable cyrupd # Linux sudo launchctl unload /Library/LaunchDaemons/com.cyrup.cyrupd.plist # macOS # Remove daemon and certificates sudo cyrupd uninstall # Clean configuration (optional) rm -rf ~/.config/{cyrupd,sweetmcp} sudo sed -i '/# SweetMCP Auto-Integration/,+5d' /etc/hosts
- API Documentation - Complete technical reference
- MCP Specification - Protocol standards
- GitHub Issues - Bug reports & features
- Discord Community - Developer chat
Dual licensed under MIT OR Apache-2.0 - choose what works for your use case.
Made with π― by Cyrup.ai