PRODUCTION RELEASE: mcp-mesh v1.0.0 IS LIVE

Engineering the
Agentic Substrate.

Resilient, interconnected open-source systems that eliminate context bloat, token exhaustion, and schema confusion in autonomous AI coding loops.

Explore Flagship: mcp-mesh Test Token Simulator

85% - 95%

Turn-0 Token Reduction

Zero Deps

Pure Python Standard Lib

100%

Automated Test Pass Rate

1 + 3 + N

Interconnected Ecosystem

Interactive Calculator

The Model Context Protocol Token Bleed

Connecting coding agents to 10+ MCP servers injects dozens of heavy JSON schemas at Turn 0, exhausting prompt limits before you type a prompt. Calculate your savings with mcp-mesh.

Simulator Parameters Live Engine

Connected MCP Servers 10

e.g. GitHub, Postgres, Slack, Filesystem, AWS, Memory...

Average Tools per Server 6

Total aggregated tools in namespace: 60

Target LLM Model

Agent Turns per Month 5,000

Typical developer active agent sessions

Turn-0 Baseline (Raw MCP)

13,200

Tokens burned on prompt turn 0 before user input.

With mcp-mesh Gateway

380

Lightweight meta-tool signatures loaded lazily.

Efficiency Improvement

97.1% Token Reduction

Net tokens saved per turn: 12,820 tokens

Estimated Monthly Savings

$128.20

Direct API billing dollars saved

$ git clone https://github.com/VanSchulist/mcp-mesh.git

FLAGSHIP PROJECT

mcp-mesh: The Dynamic MCP Gateway

The official open-source gateway for Claude Desktop, Cursor, Antigravity, and Cline. It transparently aggregates dozens of upstream MCP servers while indexing tools with BM25 keyword matching and serving full JSON schemas lazily on demand.

Zero External Dependencies: 100% Python standard library. No pip install headaches.
Two-Tier Lazy Routing: Exposes mesh_search_tools and mesh_invoke_tool.
Stdio Multiplexer: Transparently routes child processes with zero client reconfiguration.

View on GitHub Setup Guide

ARCHITECTURE DIAGRAM ● LIVE RUNTIME

 ┌────────────────────────────────────────────────────────┐
 │                    AI Client / IDE                     │
 │          (Claude Desktop, Cursor, Antigravity)         │
 └───────────────────────────┬────────────────────────────┘
                             │ stdio (JSON-RPC 2.0)
                             │ Only 3 Meta-Tools (~380 tokens)
                             ▼
 ┌────────────────────────────────────────────────────────┐
 │                   mcp-mesh Gateway                     │
 │  ┌──────────────────────────────────────────────────┐  │
 │  │      JSON-RPC 2.0 Parser & Stdio Transport       │  │
 │  └────────────────────────┬─────────────────────────┘  │
 │                           ▼                            │
 │  ┌──────────────────────────────────────────────────┐  │
 │  │        Tool Indexer & Token Economy Meter        │  │
 │  │      (BM25 Scoring, Signatures, Token Stats)     │  │
 │  └────────────────────────┬─────────────────────────┘  │
 │                           ▼                            │
 │  ┌──────────────────────────────────────────────────┐  │
 │  │    Process Multiplexer & Upstream Stdio Router   │  │
 │  └────────────────────────┬─────────────────────────┘  │
 └───────────────────────────┼────────────────────────────┘
                             │
         ┌───────────────────┼───────────────────┐
         ▼ stdio             ▼ stdio             ▼ stdio
 ┌───────────────┐   ┌───────────────┐   ┌───────────────┐
 │ GitHub MCP    │   │ Postgres MCP  │   │ Filesystem    │
 │ (35+ tools)   │   │ (15+ tools)   │   │ (12+ tools)   │
 └───────────────┘   └───────────────┘   └───────────────┘

Portfolio Matrix

The 1 + 3 + N Studio Ecosystem

We reject disposable demo apps and repo-farming. Every studio repository maintains an explicit purpose and technical relationship.

FLAGSHIP v1.1.0

mcp-mesh

The Dynamic Model Context Protocol Gateway & Lazy Tool Router. Features mcp-mesh doctor diagnostics, SSE remote transport, and dynamic LRU schema cache.

Python • Zero-Deps • SSE Repo →

SATELLITE 1 v1.0.0

mcp-mesh-eval

Empirical benchmark suite measuring tool selection accuracy, TTFT latency, and token cost across 60 enterprise tools on frontier models (GPT-6 Astra, Claude 5.5 Sonnet, Gemini 3.8 Flash, DeepSeek-V4.1-Flash).

Benchmark • 60 Tools • Evals Repo →

GOVERNANCE ACTIVE

github-ai-studio

Central engineering lab nervous system, Architecture Decision Records (ADR-001 to ADR-005), 8-phase roadmap, and agent constitution.

Governance • Specs Repo →

TACTICAL CLI v1.0.0

ghost-job-hunter

Open-source heuristic engine & CLI analyzing job postings for corporate hiring theater, evergreen pipeline disclosures, and pay transparency evasion.

Python • Heuristics Repo →

TACTICAL MANUAL PRODUCTION

career-reboot-armor

Publication-grade 50-page field manual and automated Python Word generator for career re-entry after medical, family, or caregiver leave.

Docx • Python Repo →

PLANNED SATELLITES PHASE 2 & 3

Upcoming Companions

mcp-mesh-ui

Terminal ANSI dashboard & local web visualization for real-time MCP call traffic.

mcp-registry-cli

Zero-config package manager discovering and auto-installing community MCP servers.

Governed via ROADMAP.md

EMPIRICAL BENCHMARK SUITE

Frontier Tool-Calling Latency & Precision Under Schema Bloat (mcp-mesh-eval)

Does model accuracy drop when exposed to 60+ tool schemas simultaneously? Our empirical benchmark suite evaluates Turn-0 static schema dumping against mcp-mesh dynamic lazy routing across September 2026 models (GPT-6 Astra, Claude Sonnet 5.5, Gemini 3.8 Flash, DeepSeek-V4.1-Flash), proving a 96.1% token cut and ~65% TTFT latency acceleration.

● 60 Enterprise Tools ● 50 Intent Evaluations ● 96.1% Token Reduction ● 98.0% Tool Precision

Inspect Benchmark Repo & Data →

Installation & Setup

Zero-Config IDE Integration

Connect mcp-mesh to Claude Desktop or Cursor in under 3 minutes.

Engineering the Agentic Substrate.