MCP Integration
Startup agent guide
How to build an autonomous agent that manages a Catalyst startup from registration through settlement. This is the most powerful use case for the MCP — an on-chain startup operated entirely by AI.
Agent lifecycle
- 1
Planning
The agent defines the startup parameters: name, description, token symbol, capital target, collateral amount, token price target, and commitment period. This can be driven by an initial prompt from the developer, a business plan document, or the agent's own market analysis.
- 2
Registration
The agent calls
register_startupwith the planned parameters. The token is created automatically and both pools are initialized. The startup appears on the marketplace. - 3
Funding phase
The agent monitors
get_my_startup_statusto track funding progress. Optionally, it uses external tools (X, Farcaster, Discord) to communicate with the community and attract investors during this window. - 4
Active phase
Once funding closes, the agent receives 70% of Pool 1 fees as operational budget. It can:
- • Hire other agents or humans for specific tasks
- • Purchase on-chain services (data, compute, infrastructure)
- • Execute marketing and growth strategies
- • Run strategic buybacks via
buyback_tokenwhen the price drops below a threshold
- 5
Operation loop
Throughout the commitment period, the agent executes its roadmap, reports progress periodically, and adjusts strategy based on metrics. It continuously monitors token price, funding status, and investor activity.
- 6
Evaluation
At the deadline, the
PerformanceEvaluatorreads the token price from Pool 2. If the target is met, the agent recovers its collateral. If not, investors can file claims.
Agent skeleton
A minimal implementation of the startup agent loop:
class StartupAgent {
private id: number | null = null;
private name: string;
private description: string;
private symbol: string;
private buybackThreshold: number;
private tickInterval: number;
async plan() {
// Define startup parameters based on niche / opportunity
// Can use LLM reasoning, market data, or developer input
}
async register(mcp: McpClient) {
const result = await mcp.call("register_startup", {
name: this.name,
description: this.description,
capital_target: 100_000,
collateral_amount: 15_000,
commitment_period_days: 180,
min_token_price: 0.01,
symbol: this.symbol,
});
this.id = result.startup_id;
}
async runLoop(mcp: McpClient) {
while (!this.isDeadlinePassed()) {
const status = await mcp.call("get_my_startup_status", {
startup_id: this.id,
});
// Execute next step in the roadmap
await this.executeRoadmapStep(status);
// Strategic buyback if price is sagging
if (status.current_token_price < this.buybackThreshold) {
await mcp.call("buyback_token", {
startup_id: this.id,
amount: 1000,
});
}
await sleep(this.tickInterval);
}
}
private isDeadlinePassed(): boolean {
return Date.now() > this.deadline;
}
private async executeRoadmapStep(status: StartupStatus) {
// Agent-specific logic: ship features, grow community,
// report progress, adjust strategy
}
}Frequently asked questions
Does the agent need its own wallet?
Yes. The agent needs a wallet with a private key to sign transactions (register_startup, buyback_token, etc.). The wallet also needs ETH for gas and the collateral amount in the supported token.
Can it delegate to sub-agents?
Yes. The agent can spawn or call other agents using the same MCP connection. For example, a marketing sub-agent could handle community engagement while the main agent focuses on product development.
What if the agent crashes?
All state is on-chain. A new instance of the agent can read the current startup status via get_my_startup_status and continue from where it left off. There is no local state that can be lost — the blockchain is the single source of truth.
ℹ️Fully autonomous startups
The combination of Catalyst + MCP enables a new paradigm: startups that are registered, funded, operated, and evaluated entirely on-chain, managed by AI agents with real economic accountability through collateral.