How to Build with ElevenLabs and Cursor at a Hackathon

Make Your Demo Talk Back

Judges remember the project that talks. Walk an expo floor and you pass two hundred dashboards and chatbots. The one that answers out loud in a real human voice makes judges stop and lean in. That is the whole trick, and ElevenLabs plus Cursor gets you there in an afternoon.

Below is the exact loop I run: write a spec, wire the agent to a live tool, then let it build and test itself while I babysit the spend. The counterintuitive part comes in the middle, where I make the agent write the test before the feature.

Step 0: Write the PRD Before Cursor Opens a File

AI agents write the code. They do not guess what you want. Before Cursor touches a file, write what you are building, who it is for, and what done looks like. Save it as PRD.md and AGENTS.md at the repo root so every prompt reads the same source of truth.

Let the Agent Type the Code

Nobody on a winning 2026 team types CRUD endpoints by hand. Cursor types them. Your job is the part the agent cannot do: pick the problem, define the interface, decide if the output is good, and pitch it on stage.

Time is the one thing you do not have at a hackathon. A team that hands implementation to an agent and spends the saved hours on UX, polish, and the pitch beats a team typing every line by hand. Every time.

Step 1: Check for an MCP First

An MCP (Model Context Protocol) server hands your agent a menu of typed, callable tools for a service. Instead of pasting docs and hoping, the agent sees real actions like 'generate speech' or 'clone a voice' and gets back structured data it can reason about.

ElevenLabs ships an official one at github.com/elevenlabs/elevenlabs-mcp. Drop this config into Cursor and your agent gets direct text-to-speech, voice cloning, sound effects, transcription, and the ElevenAgents platform.

{
  "mcpServers": {
    "ElevenLabs": {
      "command": "uvx",
      "args": ["elevenlabs-mcp"],
      "env": {
        "ELEVENLABS_API_KEY": "<your-api-key-here>"
      }
    }
  }
}

Step 2: No MCP? Paste the Docs URL

Some tools have no MCP server yet. The fallback: give Cursor a direct link to the official docs in the prompt. It fetches the page and uses the live API surface instead of the stale version the model remembers from training.

Here is the shape of a prompt I use to build a streaming text-to-speech endpoint with no MCP. Note the file anchor, behavior spec, constraints, and a 'Done when' section, so Cursor stops when the conditions are met instead of wandering off.

Read https://elevenlabs.io/docs/overview/intro and PRD.md > Backend Routes.

Create app/api/speak/route.ts in Next.js 15 (App Router, TypeScript).

Behavior:
- POST handler, accepts { text: string } JSON body
- Calls ElevenLabs text-to-speech with model_id "eleven_flash_v2_5"
- Voice ID and API key come from process.env (already in .env.local)
- Returns audio as a streaming audio/mpeg response

Constraints:
- Use the official @elevenlabs/elevenlabs-js SDK (install if needed)
- No retries, no caching, no logging
- Do not touch any other route or component

Done when:
- curl -X POST localhost:3000/api/speak -d '{"text":"hello"}' returns audio bytes
- The response Content-Type is audio/mpeg
- npm run build passes with no new TS errors

Step 3: Key In, Then Build and Test

Drop your API key into .env.local at the project root. The MCP server and your app code both read this file, so you never think about it again.

ELEVENLABS_API_KEY=sk_your_key_here
ELEVENLABS_VOICE_ID=JBFqnCBsd6RMkjVDRZzb
ELEVENLABS_MODEL_ID=eleven_flash_v2_5

Now hand the agent a clean task: read the PRD, build the next unfinished feature, run the test, stop only when it passes. A good agent reads the docs, scaffolds the files, installs dependencies, hits the live API, and verifies the response. You watch the loop and step in when it goes sideways.

The Trick: Make the Agent Test First

Here is the highest-leverage move I have found: flip the order. Make the agent write the test first, then build the feature that satisfies it. Classic test-driven development (en.wikipedia.org/wiki/Test-driven_development) pointed at a coding partner that happens to be an LLM.

“Never write a new piece of functionality without a failing test.”

Babysit the Spend, Then Ship

An agent in a tight loop racks up real money fast. Your Cursor subscription and your ElevenLabs credits are both on the line if you walk away from a misbehaving agent at hour 18. So do not walk away.

Get this loop running once and the voice feature stops being the scary part of your build. It becomes the thirty seconds on stage that judges actually remember.