ISSUE #001 · MCP TOOLS

GANGLIA

The routing layer
between your AI
and your code.

In a world where AI agents touch code faster than humans can think — one tool stands between your codebase and chaos.

Read the story View docs

$ cg init . && cg serve

POW!

CHAPTER 1: THE PROBLEM CHAPTER 2: THE SOLUTION CHAPTER 3: THE ARSENAL CHAPTER 4: CHOOSE YOUR SIDE

THE ORIGIN STORY

ONE COMMAND
WILL SAVE
YOUR CODEBASE.

No config. No server. No drama. Just a binary that gives your AI a brain.

$ curl -fsSL https://ganglia.dev/install.sh | sh

$ cd your-project && cg init .

$ cg serve

# Add to your MCP client config. Done.

Read the docs Quick start guide →

Ganglia hero — brain superhero with graph nodes

The AI starts editing.

"I'll refactor this function. Looks safe."

It has no idea.

DANGER!

Three hops away...

"processPayment just broke."

The blast radius is invisible.

The real problem.

! AI has no call graph. It guesses.

! No cross-session memory. Every PR starts cold.

! No impact analysis. CRITICAL paths are invisible.

! No danger zones. Auth and payments are wide open.

KA-BOOM

THIS LOOKS LIKE A JOB FOR...

GANGLIA!

57 MCP tools + Code Mode. Real call graph. One binary.

Real call graph. Instantly.

Ganglia indexes your entire codebase into a graph. Every caller. Every callee. Every route.

Impact analysis in seconds.

"code_impact shows me 12 callers before I touch a line."

Cross-session memory.

Danger zone protection.

DANGER ZONE auth · payments · env vars

Works everywhere.

Claude Code

Cursor

Windsurf

Any MCP client

Rust · TS · Python · Go

Zero config. One binary.

NEW IN 0.5.0 · CODE MODE + SEMANTIC RAG

57 TOOLS. ONE CALL.

cg_run runs sandboxed JavaScript that can call every Ganglia tool. Chain ten lookups in one turn — intermediate results stay server-side.

CODE MODE Schema tokens 10.6K → 1.5K. Typical workflow 5–10× fewer tokens.
SEMANTIC RAG Find code by meaning, not keywords. "payment retry" matches attempts_remaining.
LOCAL-FIRST Ollama default. OpenAI optional. Your code never leaves your machine.

See the tools → Read the changelog

// One cg_run call. Five tools. Zero token waste.

cg_run({
  code: `
    const hits = cg.code.semantic_search({
      query: "payment retry logic",
      top_k: 5
    });
    return hits.map(h => ({
      name: h.name,
      callers: cg.code.callers({ name: h.name })
    }));
  `
})