Axelot TAO Signer MCP

Local, non-custodial MCP signer for community-axelot trade intents.

The HS58 provider returns semantic axelot.trade-intent.v1 objects. This MCP runs on the user's machine, loads a local TAO wallet, checks the intent against local policy, reconstructs the allowlisted Subtensor call, signs locally, and optionally submits locally.

Tools

• tao_wallet_status: show local coldkey, endpoint, free balance, nonce, and policy hash.
• tao_generate_wallet: create a new sr25519 TAO coldkey mnemonic and address.
• tao_portfolio_snapshot: read local coldkey stake positions from chain state.
• tao_policy_get: return the local policy that gates execution.
• tao_trade_state: show bounded local trade state for autonomous agents.
• tao_verify_intent: validate a provider intent without signing.
• tao_dry_run_intent: show the exact reconstructed Subtensor call.
• tao_sign_trade_intent: sign locally and return signed extrinsic hex without submitting.
• tao_submit_signed_extrinsic: submit a locally signed extrinsic hex.
• tao_execute_intent: sign and submit locally to Subtensor.

Setup

Recommended npm install:

npm install -g axelot-tao-signer-mcp

Then add it to your Cursor/agent MCP config using the global binary shown below.

Fallback from the HS58 repo:

git clone https://github.com/Handshake58/HS58.git
cd HS58/providers/community-axelot/signer-mcp
npm install
npm run generate-wallet
npm run build

Use a dedicated low-value wallet. Start on Bittensor testnet or localnet before using Finney.

If the user has no TAO wallet yet, call tao_generate_wallet, put the returned TAO_COLDKEY_MNEMONIC into the local signer env, restart the signer, then run tao_wallet_status.

For CLI setup, generate a new .env with:

npm run generate-wallet

Cursor MCP Config

{
  "mcpServers": {
    "axelot-tao-signer": {
      "command": "axelot-tao-signer-mcp",
      "env": {
        "TAO_COLDKEY_MNEMONIC": "generated-or-existing-dedicated-low-value-tao-wallet",
        "SUBTENSOR_ENDPOINT": "wss://entrypoint-finney.opentensor.ai:443",
        "BITTENSOR_CHAIN": "bittensor-finney",
        "MAX_TAO_PER_TRADE": "0.01",
        "MAX_TAO_PER_DAY": "0.05",
        "MAX_TRADES_PER_DAY": "5",
        "MAX_SLIPPAGE_PCT": "1.5",
        "MIN_SECONDS_BETWEEN_TRADES": "300",
        "REQUIRE_CONFIRM": "true",
        "ALLOW_RECYCLE_ALPHA": "false",
        "ALLOWED_ACTIONS": "stake,unstake,full_unstake,move,swap",
        "ALLOWED_NETUIDS": "",
        "TRADE_STATE_PATH": "./data/trade-state.json"
      }
    }
  }
}

If you use the repo fallback instead of npm, set command to node and args to the absolute dist/server.js path.

ALLOWED_NETUIDS="" means all netuids are allowed by local policy. Use a comma-separated list such as 64,1,8 to restrict execution to specific subnets.

Execution Flow

1. Ask community-axelot for axelot/trade-plan.
2. Pass the returned intent to tao_dry_run_intent.
3. Inspect the reconstructed call and local policy verdict.
4. Call tao_trade_state to see current daily budget, active intents, and recent decisions.
5. Call tao_execute_intent with confirm:true only after user approval, unless the user explicitly configured REQUIRE_CONFIRM=false.
6. Pass the returned txHash back to axelot/monitor-trade.

If riskPolicyHash is present on an intent, it must match the local tao_policy_get hash. In guarded autopilot (REQUIRE_CONFIRM=false), tao_execute_intent also requires that the same intentId was dry-run first.

For two-step execution, call tao_sign_trade_intent first, inspect/store the signed hex locally, then call tao_submit_signed_extrinsic. Do not send signed extrinsic hex to the remote provider.

recycle_alpha requires both ALLOW_RECYCLE_ALPHA=true and confirmRecycle:true.

Autonomous Mode

Default mode is manual confirmation. If the user wants a Clawdbot-like agent to trade autonomously, they can opt in locally with REQUIRE_CONFIRM=false.

Autonomous signing is still bounded by local policy:

• MAX_TAO_PER_TRADE
• MAX_TAO_PER_DAY
• MAX_TRADES_PER_DAY
• MIN_SECONDS_BETWEEN_TRADES
• MAX_SLIPPAGE_PCT
• ALLOWED_ACTIONS
• ALLOWED_NETUIDS ("" means no netuid restriction)
• ALLOW_RECYCLE_ALPHA

The signer keeps a small bounded state file at TRADE_STATE_PATH. It stores daily budget usage, active submitted intents, and recent decisions so an autonomous agent can explain what it is in and why. It is not an append-only log.

Strategy autonomy example for agents:

{
  "autonomy": {
    "mode": "guarded_autopilot",
    "maxTaoPerTrade": 0.01,
    "maxTaoPerDay": 0.05,
    "maxTradesPerDay": 5,
    "minSecondsBetweenTrades": 300,
    "maxSlippagePct": 1.5,
    "allowedActions": ["stake", "unstake", "move", "swap"],
    "allowedNetuids": [64],
    "requireDryRun": true
  }
}

This strategy field is advisory context for agents. The actual opt-in is local: set REQUIRE_CONFIRM=false and keep strict signer limits.

Guarded autopilot still requires a matching local tao_dry_run_intent before execution. This prevents an autonomous agent from skipping call reconstruction and policy preview.

Safety Defaults

• No raw provider call data is trusted.
• Limit prices are recomputed locally from current chain pool state.
• limit_price=0 is rejected unless explicitly enabled.
• riskPolicyHash mismatches are rejected when the intent includes a hash.
• Mnemonics, keyfiles, and private keys never leave the local machine.