solon-ai

Solon-AI is a Java-based framework that integrates large language model calls, RAG knowledge bases, the MCP (Model Context Protocol), and Agent collaboration into a cohesive development experience. It supports plugging in diverse model dialects, tools, and skills to build autonomous agents, intelligent assistants, and multi-agent workflows. The MCP integration enables sharing tools, prompts, and resources across clients and servers, enabling cross-platform tool use and collaborative reasoning. Typical usage involves running the MCP-enabled server (e.g., via Docker) and connecting MCP clients to expose tools and prompts that can be invoked by chat-model-driven workflows. The framework emphasizes graph-driven orchestration for Agent reasoning, allowing you to model collaborative flows and observe or govern the execution steps.

Prerequisites:

• Java 8+ (JDK 8/11/17/21+ depending on your environment)
• Maven or Gradle for building Java projects (optional if using prebuilt binaries)
• Docker (recommended for quick setup) or a Java runtime environment if running from source

Option A: Install via Docker (recommended for quick start)

1. Install Docker on your platform.

2. Pull and run the MCP server image:

   docker run -d --name solon-ai-mcp -p 8080:8080 solon-ai/mcp-server:latest

3. Ensure the MCP endpoint is accessible at http://localhost:8080/mcp

Option B: Build from Source (requires Java build tools)

1. Install Java (JDK 8+), Maven, and Git.

2. Clone the repository:

   git clone https://github.com/opensolon/solon-ai.git cd solon-ai

3. Build the project (example with Maven):

   mvn -T 1C -DskipTests package

4. Run the generated server artifact (adjust command to your build output):

   java -jar target/solon-ai-server.jar

Prerequisites note: The SOLON-AI project targets multiple Java versions (as shown by supported JDK badges). Ensure your runtime matches the version used during build to avoid compatibility issues.

Tips and common considerations:

• MCP endpoint exposure: The endpoint path is configurable; ensure the /mcp path is accessible from MCP clients.
• Tool and Skill injection: You can augment chat workflows with Tools (for external actions) and Skills (customizable behaviors) to tailor agent capabilities.
• RAG integration: The framework supports a full pipeline from DocumentLoader to EmbeddingModel and RerankingModel for knowledge retrieval; consider indexing your knowledge base for faster responses.
• Agent coordination: Solon AI supports ReAct-style reasoning and multi-agent collaboration; use graph-driven orchestration to model complex workflows.
• Environment variables: If using Docker, you can override config locations or API keys via environment variables (e.g., SOLON_AI_CONFIG, JAVA_HOME).
• Troubleshooting MCP: Ensure client and server channels (e.g., STREAMABLE) align on endpoint paths and channel types; verify network accessibility and CORS settings if applicable.