eion

Eion’s MCP server exposes a set of session-scoped tools that let AI agents store memories, search knowledge, and manage knowledge entries within a unified memory-knowledge graph. The tools are organized into Memory tools and Knowledge tools, enabling agents to interact with conversation history and extracted knowledge via standardized tool calls. To connect, use a compatible MCP client (for example the provided Python client) and point it at the Eion MCP server. The typical workflow is to establish a session, then call tools such as add_memory, get_memory, search_memory, create_knowledge, search_knowledge, and related update/delete operations. The README provides concrete code samples for direct MCP usage with a ClientSession over a standard IO-based server (stdio) connection, as well as integrations with LangChain and Claude Desktop.

Prerequisites:

• Go 1.21+ for the Eion server (core runtime)
• Python 3.13+ for knowledge extraction services and MCP client interactions
• Docker and Docker Compose (optional, for local database services if using Docker-based PostgreSQL and Neo4j)

Installation steps:

1. Clone the repository and navigate to the project directory:

git clone <repo-url>
cd eion

2. Start optional database services (if using Docker-based stacks):

# Start PostgreSQL and Neo4j
docker-compose up -d

3. Prepare the Python environment for MCP and related services:

python3 -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate
pip install -r requirements.txt

4. Start the MCP server (Python-based MCP server integration):

# Example: run the built-in MCP server entry point
python -m internal.mcp.server

5. Run the Eion server (Go binary) as described in the Quick Start, if not already running:

# Build the Go server (if needed)
go build -o eion-server ./cmd/eion-server

# Run the server
./eion-server

6. Verify MCP connectivity by invoking a sample call (as shown in the example in the README):

# Ensure the Eion server is reachable and MCP tools are accessible
# Example to connect via the Python MCP client (see code samples in README)

Prerequisites note: Adjust paths and environment variables to suit your deployment (for example, set EION_BASE_URL for clients, and ensure databases are reachable). The README’s integration examples (Direct MCP Client, LangChain, Claude Desktop) illustrate how to wire the MCP server into common agent frameworks.

Tips and common issues:

• Ensure the EION_BASE_URL environment variable is set for MCP clients to reach the Eion server.
• When using Docker-based databases, run docker-compose up -d and verify readiness before starting the MCP server.
• The MCP tools require agent authentication (session_id, agent_id, user_id); manage credentials securely in your deployment.
• For LangChain and Claude Desktop integrations, you can translate MCP server configurations into their respective tool adapters as demonstrated in the README.
• If you encounter connectivity issues, check that the MCP server port (default 8080 per the README) is accessible from your client and that any firewall rules allow traffic.
• The Memory tools include: get_memory, add_memory, search_memory, delete_memory. The Knowledge tools include: search_knowledge, create_knowledge, update_knowledge, delete_knowledge.