CasperAgentKit

MCP-native toolkit for building autonomous Casper agents, with a working Treasury Guard reference agent.

CasperAgentKit gives an AI host a standard Model Context Protocol interface for Casper operations: read node state, inspect blocks, query global state, evaluate treasury policy, and prepare an on-chain guard action. The reference demo shows the full agent loop: perceive, decide, prepare action, and verify the receipt.

Why This Project

The buildathon rewards practical agentic applications on Casper. A plain yield router depends on testnet DeFi liquidity that may not exist. CasperAgentKit keeps the scope controllable while staying aligned with the official AI Toolkit direction: MCP, agent skills, CSPR APIs, and transaction-producing Casper components.

The submission story is:

CasperAgentKit is an MCP-native toolkit that lets AI agents read Casper state, make decisions, and execute verified testnet transactions. We demonstrate it with a Treasury Guard Agent that autonomously enforces DeFi treasury policies on Casper Testnet.

Project Structure

src/
  agent/              Treasury Guard reference agent
  casper/             JSON-RPC client and action executor adapter
  mcp/                MCP server exposing CasperAgentKit tools
contracts/            Treasury Guard contract sketch and deployment notes
dashboard/            Static demo dashboard for video recording
demo/                 Scenario input and runtime receipts
docs/                 Architecture and demo scripts

Quick Start

npm install
cp .env.example .env
npm run build
npm run demo:agent

Start the MCP server:

npm run dev:mcp

Serve the dashboard:

npm run dashboard

The dashboard runs a local API server on http://127.0.0.1:5174/. It reads demo/runs/treasury-guard-latest.json; the Run Agent button triggers a fresh local agent run.

MCP Tools

casper_get_node_status : Read node status from the configured Casper RPC endpoint.

casper_get_latest_block : Fetch the latest block.

casper_query_global_state : Query global state by key and optional path.

casper_get_account_balance : Read account information through CSPR.cloud when CSPR_CLOUD_API_TOKEN is configured.

treasury_guard_evaluate_policy : Evaluate a treasury scenario against guard policy.

treasury_guard_prepare_action : Produce a dry-run receipt or prepared live execution payload for the selected action.

Demo Agent

The Treasury Guard Agent reads demo/scenario.json, evaluates policy, prepares an action receipt, and writes the latest run to demo/runs/treasury-guard-latest.json.

The receipt includes llmAnalysis. By default the provider is local-rule-engine, so the demo works without secrets. To use a real OpenAI-compatible endpoint, set LLM_API_URL, LLM_API_KEY, and LLM_MODEL.

Default mode is dry-run. For buildathon submission, replace the executor adapter with the deployed contract call path and fill:

• CASPER_TREASURY_CONTRACT_HASH
• CASPER_SECRET_KEY_PATH
• final transaction hashes in submission_draft.md

Buildathon Proof Checklist

• Public GitHub repository
• MCP server starts locally
• Treasury Guard demo runs from CLI
• Contract address on Casper Testnet
• At least one real transaction hash
• Demo video showing agent logs, contract address, and explorer confirmation