failsafe-binance-agent
Integrates with Binance to read live BTC/USDT market data, evaluate trading intents against safety guardrails, and stage policy-checked orders for execution through Binance's confirmation flow.
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@failsafe-binance-agentEvaluate a market buy of 0.5 BTC with a 2% risk limit."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Failsafe
Failsafe is an execution-firewall prototype for trading agents, built for the Binance Agent OS Mini Hackathon. Any agent can submit a proposed action; Failsafe locally evaluates supplied evidence against versioned policy and either blocks the proposal or emits a deterministic decision receipt for human review.
Why it exists
Most trading agents keep planning and policy inside the same prompt. That lets a hallucinating, compromised, or overly eager planner rationalize its own exception. Failsafe models a deterministic policy boundary between the planning agent and Binance MCP.
Intended production flow:
Any agent → proposed action → Failsafe firewall → human approval → Binance MCP
↓
deterministic decision receiptRelated MCP server: Aegis
Demo flow
The user describes an intent with risk constraints.
The dashboard displays live BTC/USDT market data from Binance, or replays a timestamped MCP snapshot in proof mode.
It evaluates momentum, freshness, single-order size, rolling exposure, balance evidence, and liquidity.
Missing evidence fails closed; it is never rendered as a pass.
Given supplied rolling history, it detects split-order evasion while retries with the same proposal ID remain idempotent. Production enforcement requires a durable server-side ledger.
The firewall emits a deterministic SHA-256 content fingerprint and decision receipt containing its policy version, evidence, checks, reasons, and execution state.
The public demo intentionally stops before live execution. In a production adapter, only a fully evidenced receipt would be eligible for Binance's own human-confirmation boundary.
Binance Agent OS integration
The default dashboard reads live ticker and order-book data from Binance's public API; the
?proof=mcpview replays the timestamped genuine Binance MCP evidence and labels it as a recorded snapshot.The companion agent policy in
agent/FAILSAFE.mdis designed to run in a Binance MCP-connected AI client.A production execution adapter should follow Binance MCP semantics: least-privilege scopes, no withdrawal scope, and confirmation before each trading action.
The interface exposes
evaluate_trade_proposalthrough WebMCP, allowing any browser-connected agent to request the same deterministic evaluation without receiving trade authority. It evaluates agent-supplied evidence and does not attest to its provenance.The firewall core is independent of React and covered by adversarial Node tests.
Prototype boundary
The public site evaluates proposals locally and deliberately has no Binance Trade scope, account access, order endpoint, or server-side receipt ledger. Its receipt ID is a reproducible SHA-256 content fingerprint, not a server signature or an authorization token. A production deployment must persist decisions and enforce an authorization check in the only adapter that has Binance Trade capability; an agent must not be able to call Binance directly around that adapter.
See the timestamped verified Binance MCP read, where Failsafe used genuine market data and blocked an order whose positive-momentum condition was not met.
Run locally
npm install
npm test
npm run devSafety
This is a hackathon prototype, not financial advice. The hosted demo does not place orders or access account funds.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Connectors
Advisory policy preflight for AI-agent spend requests; never executes payments or accesses wallets.
Pre-execution governance for AI agents. Deterministic PASS/FAIL/REVIEW verdicts, replayable proof.
MCP enforcement layer that intercepts AI agent actions and blocks rule violations before execution.
Supervised API-write gateway for AI agents with policy, human approval and execution receipts.
1
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables risk-gated AI trading on Binance by researching market data, enforcing policy checks, and building executable MCP order requests that require user confirmation before the official Binance MCP server executes them.MIT
- FlicenseNot gradedqualityCmaintenanceEnables AI agents to trade on Binance through a governed MCP proxy that enforces configurable policies, requires a recorded rationale before orders, detects prompt injection, logs all actions in a tamper-evident audit trail, and blocks execution until a human approves.-
- FlicenseAqualityCmaintenanceEnables Binance Agent OS to perform pre-trade risk assessment, parse natural language trade intents, benchmark yield arbitrage opportunities, and execute hedged trades with automated safety gating.4-
- AlicenseNot gradedqualityBmaintenanceEnables budget-bound Binance trading workflows with preview and human approval, durable recovery, receipts, and reconciliation of interrupted orders without resubmission.2MIT