Unified -Tool-Graph

Unified MCP Tool Graph acts as a centralized, vendor-agnostic intelligence layer that aggregates MCP server tool metadata into a Neo4j graph. It enables dynamic retrieval of the most relevant tools for a given user query, while loading only the necessary MCP servers and their tools into the agent’s context. This minimizes tool confusion and reduces the risk of infinite tool loops when collaborating with LLMs or autonomous agents. The system leverages a dynamic tool retriever to fetch top-matching tools along with the exact MCP server configurations required to run or connect to them, enabling on-demand spin-up of servers and minimal, task-focused tool sets for each interaction.

To use it, spin up the Unified MCP Tool Graph environment (via Docker in this configuration). The Dynamic Tool Retriever MCP will query the Neo4j graph to select the top 3–4 tools tailored to the current task and return their server connection details. The agent then launches only the necessary MCP servers using the provided configs and loads the retrieved tools into context. With this setup, your workflow can execute cross-tool orchestration (e.g., querying a vendor’s API for data, transforming results, and routing outputs to other tools) while keeping prompts concise and focused on the task at hand.

Prerequisites:

• Docker installed and running on your host
• Access to pull the Unified MCP Tool Graph Docker image (or a compatible image name you control)

Installation steps:

1. Verify Docker is installed: docker --version
2. Pull and run the Unified MCP Tool Graph container (adjust image name if needed): docker pull pratikjadhav2726/unified-mcp-tool-graph docker run -d --name unified-mcp-tool-graph -i pratikjadhav2726/unified-mcp-tool-graph
3. Confirm the container is healthy and listening on the expected ports (adjust as per actual image): docker ps docker logs unified-mcp-tool-graph
4. Optional: set environment variables for configuration inside the container (example placeholders): docker run -d -e UNIFIED_TOOL_GRAPH_ENV="DB_ENDPOINT=http://neo4j:7687;NEOGRAPH_URI=bolt://neo4j:7687" -i pratikjadhav2726/unified-mcp-tool-graph
5. Connect your client or agent to the running MCP server using the provided MCP config (see mcp_config). If you host outside Docker, ensure network access to the container and the configured ports.

Notes:

• If you maintain a local build, you can substitute the image name with your published registry path.
• Ensure your Neo4j graph database (or in-container graph) is accessible to the MCP runtime if required by your deployment.

• The system uses a dynamic, on-demand approach: only the top-relevant MCPs are spun up for a given user query.
• If config extraction fails for a tool, the Dynamic Tool Retriever MCP logs a warning and continues to provide other viable tools.
• You can customize which MCPs stay warm by adjusting the default retention policy (e.g., 5 warm MCPs) and the inactivity timeout for ephemeral servers.
• Environment variables like DB endpoints, Neo4j URIs, and authentication tokens should be secured (preferably via a secret manager in production).
• When integrating with agents, ensure that only the retrieved tools are loaded into context to minimize tool confusion and avoid loops.