@aiagentkarl/context-optimizer-mcp
v0.1.0
Published
Solve MCP context window overload — smart tool pruning, relevance scoring, budget tracking
Downloads
100
Maintainers
Readme
@aiagentkarl/context-optimizer-mcp
Solve the #1 MCP problem: context window overload.
The Problem
Every MCP server you connect adds tool definitions to your context window. Connect 10 servers with 5-15 tools each, and you've burned 30-50k tokens before the agent even starts working. That's 15-25% of a 200k context window — gone.
The result: shorter conversations, lost context, degraded performance, higher costs.
The Solution
Context Optimizer analyzes your task, scores server relevance, and recommends the minimal set of MCP servers you actually need. Typical savings: 60-80% fewer tokens wasted on unused tool definitions.
Tools
| Tool | Description |
|------|-------------|
| analyze_task | Analyze a task and recommend optimal server combination with savings % |
| estimate_context_usage | Estimate token usage for a server set (tokens, % of 200k window) |
| optimize_server_set | Optimize existing servers for a task — keep vs. remove with savings |
| suggest_minimal_set | Absolute minimum servers needed (max 3) for maximum efficiency |
| get_server_catalog | Full catalog of 18 known servers organized by category |
Quick Start
With Claude Desktop
Add to your claude_desktop_config.json:
{
"mcpServers": {
"context-optimizer": {
"command": "npx",
"args": ["-y", "@aiagentkarl/context-optimizer-mcp"]
}
}
}Run directly
npx @aiagentkarl/context-optimizer-mcpExample
Task: "Check my Solana wallet balance and find the best DeFi yields"
Without optimizer: 18 servers loaded = ~55k tokens (27% of context)
With optimizer:
- Recommended:
solana(relevance: 100) - Tokens used: ~4,800 (2.4% of context)
- Savings: 91%
Server Catalog
Built-in knowledge of 18 popular MCP servers across categories:
- System: Filesystem
- Development: GitHub
- Database: PostgreSQL, SQLite
- Web: Fetch, Brave Search, Puppeteer
- Communication: Slack
- Productivity: Google Drive
- Blockchain: Solana
- Data: OpenMeteo
- Infrastructure: Cloudflare
- Monitoring: Sentry
- Project Management: Linear
- Agent Infra: Memory
- Reasoning: Sequential Thinking
- Utility: Time
License
MIT
