jgravelle

jgravelle / jmunch-mcp

Public

Handle-ifying MCP proxy: slash token bills on chatty MCP servers via content-aware backends and local munch verbs

11
4
100% credibility
Found Apr 24, 2026 at 11 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

A proxy tool that wraps data-fetching services for AI agents, replacing large responses with compact, queryable handles to drastically cut token usage.

How It Works

1
🔍 Discover token saver

You hear about a simple tool that makes your AI chats cheaper and faster by smartly handling big data dumps.

2
📦 Add it easily

You grab the tool onto your computer in moments, no hassle.

3
⚙️ Quick setup magic

Run a friendly scan to find your data tools, pick which ones to boost, and it creates perfect wrappers for them.

4
🚀 Swap in the booster

Tell your AI app to use the boosted versions instead — one tiny change.

5
💬 Chat smarter

Now when AI pulls huge info, it gives a tiny summary with a magic handle; ask peek, slice, or search to dive in without wasting space.

6
📊 See the wins

Open the dashboard to watch tokens saved skyrocket — 90% less cost, quicker answers.

🎉 AI supercharged

Your AI tools run leaner, faster, and way cheaper, handling big data like a pro every time.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 11 to 11 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is jmunch-mcp?

jmunch-mcp is a Python proxy for MCP servers and OpenAI/Anthropic APIs that slashes token bills on chatty MCP servers by handle-ifying large tool responses into content-aware backends. Instead of dumping fat payloads like GitHub issues or Firecrawl scrapes into the model's context, it swaps them for compact handles plus summaries, letting agents drill down with local munch verbs like peek, slice, search, and aggregate. Run it via CLI with `jmunch-mcp --config file.toml` or gateway mode by pointing apps at `http://127.0.0.1:7879`.

Why is it gaining traction?

It delivers 88-99% token savings and faster response times in benchmarks against real MCP servers like GitHub and Firecrawl, with no app changes needed—just swap the base URL or config. The `init` command auto-scans client configs (Claude, Cursor) and generates wrappers, while a built-in dashboard tracks savings per upstream. Universal verbs work across backends, making it dead simple for any tool-heavy workflow.

Who should use this?

AI agent builders hitting token limits on verbose MCP tools like GitHub PR lists or web scrapes in Claude Desktop/Cursor. Teams proxying OpenAI-compatible apps (LangChain, Aider) through Anthropic who want to trim bills without rewriting prompts or pagination logic.

Verdict

Grab it if you're on MCP—benchmarks prove it pays off fast, and MIT licensing lets you embed freely. At 11 stars and 1.0% credibility, it's early (v0.2) with thin docs, so test in a side project first.

(187 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.