hackquest logo

InjPilot MCP

Safety-first AI command center for Injective: live MCP reads, local risk simulation, portfolio safety scans, and preview-only execution with no signing or broadcast.

Videos

Project image 1
Project image 2
Project image 3
Project image 4

Tech Stack

Next
Web3
React
Node
Python

Description

InjPilot MCP is a safety-first AI command center for Injective. It uses the official Injective MCP Server for live read-only market and account data, Gemini/Copilot explanations, local risk simulation, portfolio safety scans, and preview-only execution reports.

The project solves a real DeFi safety problem: AI agents can suggest trading actions, but users need a trusted inspection layer before any wallet signs or broadcasts a transaction. InjPilot MCP shows what a safer AI-to-onchain workflow can look like.

Users can:

  • Read live BTC, ETH, and INJ perp market data through Injective MCP.

  • Simulate trades such as Simulate $50 ETH long at 5x.

  • Inspect public Injective addresses through read-only account balance scans.

  • Prepare an execution preview for a trade_open action without calling trade_open.

  • View a transparent MCP audit trail showing allowed reads, prepared previews, and blocked execution paths.

InjPilot intentionally does not connect wallets, request private keys, sign transactions, or broadcast orders. The safety layer only allows market_listmarket_price, and account_balances. Write and execution tools such as trade_opentransfer_sendwallet_importauthz_grantbridge_debridge_send, and evm_broadcast are blocked by default.

The result is a deployed safety cockpit for Injective: live MCP reads, local risk math, Gemini/Copilot explanations, auditable tool calls, and execution previews with signing permanently disabled.

Progress During Hackathon

During the hackathon, I built and deployed InjPilot MCP as a full-stack Next.js application connected to the official Injective MCP Server.

Key progress completed:

  • Verified the official Injective MCP server locally using MCP Inspector.

  • Built a Next.js TypeScript frontend with a multi-page command-center UI.

  • Added live MCP integration for read-only tools: market_listmarket_price, and account_balances.

  • Enabled live MCP support on Vercel by vendoring and building the MCP server inside the deployed app.

  • Added Gemini/Copilot explanation support with safe local fallback.

  • Built Market Intelligence for BTC, ETH, and INJ market reads.

  • Built local risk simulation showing margin, notional exposure, estimated size, +1%/-1% PnL, and rough liquidation.

  • Built Portfolio Safety Scan for read-only public Injective address checks.

  • Built Execution Preview Report that prepares inert trade_open parameters but never signs or broadcasts.

  • Built MCP Audit Trail showing read-only tools, preview-only tools, and blocked execution tools.

  • Added a guided product flow to walk users through the main product story.

  • Added safety documentation, architecture documentation, README screenshots, and Playwright product-flow tests.

  • Hardened the production health endpoint so it does not expose debug paths in production.

  • Confirmed deployed Vercel flows pass: live MCP market read, ETH simulation, INJ execution preview, invalid address validation, and audit trail.


Team Leader
RRitesh S
Project Link
Deploy Ecosystem
InjectiveInjective
Sector
AIDeFiInfra