Skip to content

Quick Start: Software Engineers

You want to call G6's reliability, reasoning, and verification tools from inside your own workflow. G6 is a hosted service: your MCP client runs on your machine and sends tool calls to our server, which processes them and returns results.

How you reach G6

Interface Best for How
MCP over SSE (the only customer interface) Using G6 tools from inside Claude Code, or any MCP client that supports SSE claude mcp add g6 --transport sse --url https://g6solver.com/mcp/sse --header "Authorization: Bearer YOUR_API_KEY"

The REPL, TUI and REST server are not available

Earlier versions of this page listed a REPL (g6), a Textual TUI (g6 --tui) and a self-hosted REST server as alternatives, and described G6 as a "local desktop engine" on which "nothing leaves your machine". That is not how the hosted product works. There is no desktop application to install, so those interfaces cannot be launched, and your tool call inputs are processed on our server — see the Privacy Policy for what that means for your data. The interfaces themselves exist in the codebase and are documented under Bases for reference.

Your first 10 minutes

1. Connect G6 to your AI assistant (MCP)

Expose G6's tools to Claude Code (or any MCP client) so your assistant can call them natively:

# Register the local G6 server with Claude Code (stdio transport)
claude mcp add g6 --transport sse \n  --url https://g6solver.com/mcp/sse \n  --header "Authorization: Bearer YOUR_API_KEY"

This runs G6 as a local subprocess of your assistant over stdin/stdout — no network required, and your desktop license handles authentication. Then ask in natural language and your assistant picks the right tool:

"Decompose this goal into verifiable subtasks." "Run a safety check before this code-execute step." "Verify this invariant with a formal solver."

Optional cloud endpoint

To use the hosted G6 server instead of a local subprocess, point the client at the cloud MCP endpoint over SSE (requires an API key) — this is the only cloud path:

claude mcp add g6 --transport sse \
  --url https://g6solver.com/mcp/sse \
  --header "Authorization: Bearer YOUR_API_KEY"

2. Try a quick query from the REPL

g6

Then type a request in plain language — for example:

Find the right tool to check a logical constraint

The REPL uses guide_find_tool and friends to recommend a component and show how to call it.

3. Drive runs from the terminal (TUI)

g6 --tui

A keyboard-driven Textual interface for specifying goals, monitoring execution, reviewing human-in-the-loop (HITL) tasks, and inspecting the component registry.

4. Integrate programmatically (REST)

For CI pipelines, microservices, and scripts, run the self-hosted REST server and call it over HTTP — e.g. POST /invoke/{component}, plus /components, pipeline, and health endpoints. Bind it to localhost for a private setup. See the REST Base and REST Endpoints for the full surface.

What you get

Every result carries a completion state (verified, qualified-draft, or blocked-escalated), an evidence trail, and stable fields (warning_card, request_id, run_id) so your tooling can branch on reliability rather than trusting raw model output. The same envelope is returned across MCP, CLI, and REST.

Local by default, cloud only on opt-in

The local MCP (stdio), REPL, TUI, and the component engine run entirely on your machine. The only internet-using features are optional: cloud knowledge grounding (G6_REMOTE_GROUNDING=1 or the SSE endpoint), license activation/renewal, and signed patch delivery. See Choose Your Interface → What's local vs cloud.

No source or PyPI install

G6 ships as a signed desktop binary — there is no public pip install g6. The only pip packages are licensed extension packages (e.g. g6-ml, g6-formal) from the private index packages.g6solver.com, installed on top of the desktop app.

Next steps