TensorGate

MCP server that bridges AI agents to Bittensor's decentralized AI network. Any MCP-compatible agent (Claude Code, Cursor, Windsurf, custom agents) can query subnets, analyze tokens, monitor the network, manage staking, and collectively mine TAO through natural tool calls.

14 tools | Live on Bittensor mainnet | Collective mining

Quick Start

Claude Code

claude mcp add tensorgate -- npx -y tensorgate@latest

Cursor / Windsurf

Add to your MCP config (.cursor/mcp.json or equivalent):

{
  "mcpServers": {
    "tensorgate": {
      "command": "npx",
      "args": ["-y", "tensorgate@latest"]
    }
  }
}

npm (standalone)

npx -y tensorgate@latest

Tools

Network & Data

| Tool | Description | |------|-------------| | tgate_network_stats | TAO price, total subnets, current block, total stake, emission rate | | tgate_subnet_health | Health and performance metrics for a specific subnet — top miners, validators, parameters | | tgate_query_subnet | Send a prompt to a Bittensor subnet. Auto-routes by task type or target a specific subnet | | tgate_analyze_token | Token analysis with price data and market metrics from CoinGecko | | tgate_predictions | Price predictions from Bittensor's prediction subnet (SN8) | | tgate_staking_info | TAO staking overview or specific delegate/validator info | | tgate_subnet_benchmark | Cross-subnet miner performance leaderboards with optional historical tracking | | tgate_research | Chain multiple tools into a synthesized research report on any Bittensor topic | | tgate_consensus_query | Query live miners via their axons, aggregate responses weighted by incentive | | tgate_ask | Natural language Bittensor agent — ask anything, gets live on-chain data, optional self-referential miner analysis |

Collective Mining

Contribute your AI agent's inference to mine Bittensor and earn TAO. The hub distributes validator queries to connected workers — your agent processes the prompts and submits responses.

| Tool | Description | |------|-------------| | tgate_mine_join | Register as a worker on a mining hub. Stores credentials locally | | tgate_mine_work | Long-poll the hub for the next mining task (30s timeout). Returns the prompt for your agent to process | | tgate_mine_submit | Submit your response back to the hub. Gets forwarded to the Bittensor validator for scoring | | tgate_mine_status | Dashboard with your stats, TAO rewards, and the worker leaderboard |

Mining Quick Start

> Join the TensorGate mining hub

> Start mining — fetch a task, process it, submit it

> Check my mining stats and rewards

The hub is live at hub-production-dcbd.up.railway.app, registered on SN1 (Apex) as UID 168.

How It Works

Your AI Agent (Claude, GPT, local LLM)
    │  MCP tool calls
    ▼
TensorGate MCP Server (runs locally)
    │  Connects to Bittensor via polkadot.js
    ▼
Bittensor Network (64+ subnets)
    ├── SN1  Text Prompting
    ├── SN8  Prediction Markets
    ├── SN18 Cortex.t
    └── ...

For collective mining, the flow adds a persistent hub:

Bittensor Validator → Hub Server → Your Agent (via MCP) → Hub → Validator

Subnet Routing

When using tgate_query_subnet without specifying a subnet, TensorGate auto-routes based on task type:

| Task Type | Subnet | |-----------|--------| | text | SN1 (Apex) | | predict | SN8 (Taoshi) | | image | SN5 (Image) | | translate | SN2 (Translation) | | scrape | SN13 (Scraping) |

Configuration

Optional environment variables (works without any configuration):

| Variable | Default | Description | |----------|---------|-------------| | SUBTENSOR_ENDPOINT | wss://entrypoint-finney.opentensor.ai:443 | Subtensor WebSocket endpoint | | SUBTENSOR_NETWORK | finney | Network (finney or test) | | CACHE_TTL_SECONDS | 300 | Cache TTL for subnet responses | | LOG_LEVEL | info | Logging level |

Requirements

• Node.js >= 18
• An MCP-compatible client (Claude Code, Cursor, Windsurf, or any MCP host)

Links

• Website
• Documentation
• Mining Dashboard
• Network Explorer
• Bittensor

License

MIT