Cut LLM Token Costs by Routing Code Context
A standardized interface for AI assistants to navigate, understand, and interact with your repository's engineering context.
The Cost of Codebase Blindness:
The Token Tax
AI coding agents are incredibly powerful, but they operate
blind. When asked a simple question like
"what functions call ProcessPayment?", agents typically traverse your repository file-by-file.
They grep across files, read entire modules, and follow imports sequentially. This raw document exploration can easily burn 40,000+ tokens on a single simple query. Over a development session, this "token tax" leads to runaway API costs, slow response times, and frequent context limit exhaustion.
A Structural Map for
AI Coding Agents
The Context Router MCP solves token bloat by mapping your repository's skeletal structure, classes, functions, imports, exports, and relationships. It exposes this schema directly to your AI agent, allowing it to navigate code boundaries instantly.
Structural Navigation
Instead of guessing filenames or reading entire directories, the agent queries codebase skeletons to trace definitions, imports, and cross-references.
Context Compression
The agent retrieves only the relevant code signatures and boundaries, stripping out unnecessary implementation details to keep payloads compact and fast.
Penqwin Context Router vs. File-by-File Search
Transitioning from blind file crawling to structural context routing reduces token usage by up to 99%.
| Query Type | Penqwin Context Router | File-by-File Search | Savings |
|---|---|---|---|
| Find function by pattern | ~200 tokens | ~45,000 tokens | 225x savings |
| List all HTTP routes | ~400 tokens | ~62,000 tokens | 155x savings |
| Architecture overview | ~1,500 tokens | ~100,000 tokens | 67x savings |
| Total Session Average | ~2,100 tokens | ~207,000 tokens | Up to 99% reduction |
IDE Setup &
Configuration
Connect the Penqwin Context Router to your favorite AI assistant
in seconds using the standard Model Context Protocol (MCP).
Your PENQWIN_ORG_ID and PENQWIN_API_KEY can be retrieved from the app.penqwin.com/settings → integrations tab.
Add the server setup to your Claude Desktop configuration
file.
Path: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"penqwin": {
"command": "npx",
"args": ["-y", "@penqwin/mcp"],
"env": {
"PENQWIN_API_KEY": "YOUR_API_KEY",
"PENQWIN_ORG_ID": "YOUR_ORG_ID",
"PENQWIN_REPO": "owner/repo_name",
"PENQWIN_API_URL": "https://app.penqwin.com"
}
}
}
}
Designed for
Engineering Teams
The Context Router MCP is engineered for speed and footprint efficiency, easily scaling to compile and route structural blueprints for enterprise-grade repositories.