NeuroMesh

Ship less context. Ship the right code.

Local-first MCP context engine for Cursor, VS Code, Claude, Codex, and every MCP client. NeuroMesh indexes your repo into a graph, routes your prompt to the right symbols, and sends a folded evidence packet — not thousand-line file dumps.

Cursor · VS Code · Claude · Codex · OpenCode · MiMo CLI · Antigravity · Kilo · Trae · Windsurf · Zed

The pain · Fold · Galaxy · Measured · Install · Connect · Docs · Site

The pain

You ask a simple question in a large project. The editor copies two or three thousand-line files and ships them to the model.

What you pay for:

1. Tokens you never needed — dollar cost on every turn
2. Seconds of fake loading while the window fills with helpers you will not touch
3. Lost in the middle — the model drowns in unrelated bodies and invents bugs

Today’s workarounds all leak in a different place:

NeuroMesh is the missing step: route first, then fold. The graph is for finding the path. The packet is what the model actually reads.

Don’t delete. Fold.

Don’t delete the extra code. Fold it.

How does nature pack two metres of DNA into a nucleus without deleting a single letter?
Not by throwing genes away — by folding.

Nature does not delete DNA to fit a nucleus. It supercoils.

NeuroMesh treats the syntax tree like a genetic strand in RAM:

• Functions you need stay expressed (exons) — real body, real lines
• The rest collapse to a one-line reversible intron:

/* [neuromesh:fold:fold_unused_helper_1 | 12 lines folded | fn unused_helper()] */

The agent still sees the shape of the file — signatures, imports, neighbors — without paying for every private helper. When a folded body is required, neuromesh_expand_fold unsplices it from a registry in memory. Nothing was deleted. Nothing needs a second grep of the disk.

Structure stays. Tokens sleep. Wake a fold when you need it.

Why it feels different

Curious about the biology metaphor? docs/nature.md

Install

Pre-built binary (no Rust required). v0.9.0 defaults to engine: fast — instant graph index, no ONNX warm at startup.

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/pinoox/neuromesh/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/pinoox/neuromesh/main/install.ps1 | iex

Then from your project root:

neuromesh doctor          # verify binary and workspace
neuromesh connect         # write MCP configs (Cursor, VS Code, Claude, …)
neuromesh index           # build the graph (<30s typical)

Restart your IDE so MCP picks up the new server. Re-run the installer to update — then neuromesh -V should show v0.9.0.

Hybrid/deep embeddings (neuromesh install embed minilm), CBM proxy, monitor port: docs/configuration.md.

Connect

NeuroMesh speaks MCP over stdio — what your IDE launches in the background.

neuromesh connect --global --agent-rules   # recommended once per machine

That registers the server and copies the agent rule so the IDE actually calls NeuroMesh instead of raw Read / Grep.

Manual (when neuromesh is on PATH) — paste into ~/.cursor/mcp.json:

{
  "mcpServers": {
    "neuromesh": {
      "command": "neuromesh",
      "args": ["mcp"]
    }
  }
}

Per-client paths: docs/mcp.md.

Agent loop

Pass the user task as written — any language. Default engine: fast: graph + server-assisted concept expansion; no keyword tables.

get_context_packet(query / task_description / prompt / task)
  → check coverage.claim and retrieval.resolution_tier
  → neuromesh_search_symbols or neuromesh_expand_gap if seeds missed
  → neuromesh_expand_fold when a folded body is required
  → neuromesh_trace for callers and blast radius
  → neuromesh_record_feedback after a successful edit

Teach every IDE: docs/agent-guide.md · Cursor template: docs/agent-rule.mdc.

3D Neural Galaxy

neuromesh monitor is a live map of your project: packages at a glance, then the file graph, then symbols inside a module.

Constellation — packages and subsystems

3D galaxy — files and call/import links

Module zoom — files and symbols in one area of your codebase

Default URL: http://127.0.0.1:8765 · neuromesh monitor · port: neuromesh port

Tools (MCP)

Full reference: docs/mcp.md.

Everyday CLI

neuromesh index              # refresh graph after large changes
neuromesh status             # node / edge counts
neuromesh monitor            # 3D graph UI (see above)
neuromesh doctor --engine    # show retrieval preset
neuromesh config engine hybrid   # opt in to semantic search (needs embed install)

Command reference: docs/cli.md.

Languages

Rust, TypeScript, Python, Go, Java, Kotlin, PHP, C#, Dart, Swift, Ruby, and more via tree-sitter. Framework overlays for Laravel, Django, Next, Vue, Axum, Rails, Flutter, and others. Details: docs/architecture.md.

What we actually measured

Savings are per task, after folding — not a marketing average. Run neuromesh eval on your own repo to see your numbers.

Example from a 650k-token monorepo (release v0.9.0, default engine: fast):

Index on that project: 340 files · 552 ms. Methodology and multilingual holdout: docs/quality.md.

Documentation

MIT · LICENSE