Skip to content

Setup

Create an account, get an API key, and point your agent at the hosted endpoint — in a few minutes. G6 runs on G6Solver's servers; you reach it over the network.

Step 1: Create an account and get an API key

Sign up at g6solver.com, then open Dashboard → API Keys and create a key. A 30-day free trial is available; see Pricing for plan details.

There is nothing to download

G6 is a hosted service. There is no installer, no desktop build, and no source checkout to obtain — the LICENSE grants access solely through the hosted service. pip install g6 does not install this product, and no public repository serves it.

Step 2: Choose a plan

Your subscription tier determines which tools and rate limits your key reaches.

Tier Tools Rate Limit Price
Free Trial 30 core tools 10 req/min 30 days free
Researcher 49 tools (Core + Navigator + Basic Extras) 100 req/min $20 AUD/mo
Builder Full tool access + dashboard + multi-device 500 req/min $50 AUD/mo

Step 3: Connect Claude Code to the hosted endpoint

Register G6 as an MCP server over SSE. This is the supported path.

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

{
  "mcpServers": {
    "g6": {
      "transport": "sse",
      "url": "https://g6solver.com/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Get your API key from the Dashboard → API Keys section.

Keep your key secret

Never commit API keys to version control. Store them in environment variables or a secret manager.

Grounding and entailment models

The grounding and document-verification features depend on a sentence-embedding model and an NLI cross-encoder. These are provisioned on the server, not on your machine. If a model is absent, the affected feature abstains (it will not silently guess).

Step 4: Try your first tools

In Claude Code, describe what you want — you don't need to know tool names:

Check the G6 system status

Calls the system_status tool to verify your connection.

Use G6 to decompose: Analyze this codebase and identify potential memory leaks

G6 breaks the goal into subtasks using the goal_engine component (decompose_goal).

Use the G6 run_safety_pipeline tool to run a safety verification pipeline

Pipelines compose multiple components into sequential or parallel flows.

List all available G6 components

Calls list_components to return metadata for every registered component.

What runs where

Your agent runs on your machine. Everything G6 does runs on G6Solver's servers.

flowchart TD
    subgraph Yours["Your machine"]
        UI["Claude Code / your MCP client / your HTTP client"]
    end
    subgraph Hosted["G6Solver's servers"]
        Engine["G6 engine"]
        Reg["Component registry"]
        Ground["Knowledge grounding corpus"]
        LLM["Model routing (Ollama / OpenRouter)"]
    end
    UI -- "MCP over SSE, or REST" --> Engine
    Engine --> Reg
    Engine --> Ground
    Engine --> LLM

Only what you explicitly pass to a tool leaves your machine. Model routing and spend limits are enforced server-side.

Troubleshooting

If a tool reports a missing capability, that cluster is outside your plan — components outside your plan's clusters return a clear diagnostic message rather than crashing. See Troubleshooting.

Next steps