stevesolun/ctx

Not an Amazon-style catalog or marketplace. ctx is a recommendation layer: bring your org tools or use the shipped graph to load the right skills, agents, MCPs, and harnesses only for the current dev window, cutting token bills and local compute waste: 79,958-node LLM-wiki graph, 68,494 skills, 467 agents, 10,790 MCPs, 207 harnesses.

What it solves

ctx addresses the problem of "context budget" and "skill rot" in AI development. When developers have access to thousands of potential skills, agents, and Model Context Protocol (MCP) servers, loading all of them into an LLM's context window wastes tokens, degrades response quality, and consumes unnecessary CPU/GPU resources. ctx prevents this by recommending only the smallest, most relevant bundle of tools for the specific task at hand.

How it works

ctx acts as a recommendation layer that utilizes a massive knowledge graph (containing nearly 80,000 nodes and 1.7 million edges) to map relationships between skills, agents, and MCP servers. It monitors the user's current project and development window to identify the most useful tools.

Depending on the setup, it can integrate directly with Claude Code via hooks or work with local/API models through separate harness and adapter flows (such as LoopFlow). It provides a CLI for scanning repositories, recommending bundles, and managing the installation of "harnesses" (execution environments) and other entities.

Who it’s for

It is designed for developers and organization-wide teams using LLMs, AI agents, and MCP servers who need a way to dynamically manage which tools are active in their AI's context to optimize performance and cost.

Highlights

  • Massive Recommendation Graph: Leverages a pre-built graph of over 79,000 nodes to score and suggest the best tools.
  • Context Optimization: Reduces token burn and compute waste by loading only the 10–15 most relevant tools per session.
  • Harness Management: Includes a dedicated system for recommending and installing harnesses for local or API-based models.
  • Integrated Monitoring: Features a ctx-monitor dashboard for visualizing loaded skills, viewing the knowledge graph, and tracking token usage.
  • Detailed Quality Scoring: Provides a four-signal quality score for every skill to help users evaluate utility.