Most hackathon entries are decided before the demo. They're decided by what the team chose to build in the first two days, and by whether the final pitch can show something real.
Both come down to information: who already built this, what happened to them, how big the market is this month, what users complain about, and which tools actually ship. Most of it isn't on the free web, and the part that is, every other team has already read.
Since October 1st, Frames is built into Colosseum Copilot, the research skill Colosseum gives every hackathon builder. Ask Copilot something that needs current evidence from outside its archive, and it can buy the answer through Frames, with your approval before every charge.
This post covers where the edge in a hackathon actually comes from, how Frames gives it to your coding agent, and how to use it inside Copilot while the Crypto World's Fair is still open.
Where crypto hackathons are actually won
A hackathon has three moments that decide the result, and only one of them involves code.
Picking the problem. Colosseum has 8,286 submissions on file from five hackathons, and every obvious idea is in there several times over. Copilot's own data shows payments projects doubled as a share of submissions, from 4.1% at Renaissance to 8.4% at Frontier. If you're building payments this month, dozens of teams had your idea first. The edge is knowing exactly what they shipped, where they stopped, and which of them are still alive.
Proving it in the pitch. A deck that says the market is big loses to a deck that says how many wallets did the thing last month and links the source. Judges can't verify claims live, so sourced numbers do the convincing for you.
Running the demo on real data. A product that runs on live prices, real wallet labels or a real order book is a different product from one running on a mocked JSON file, and judges can tell.
All three depend on data that is current, specific and, mostly, paid for.
Why the free web won't give your team an edge
Every team's AI reads the same web. Ask any chatbot for gaps in Solana DeFi and you get the same five ideas as the other 9,000 builders registered for the World's Fair who asked it this month.
If the information is free, every other team has it too.
The data that separates a pitch from the pack lives somewhere else: funding databases, on-chain analytics, token and market data, full forum threads and hiring pages, in products like Nansen, Messari, CoinGecko and Dune. Each sits behind an API key, a subscription and a credit card, and nobody buys six subscriptions for a four-week hackathon.
Those paid sources are exactly what Frames reaches.
How Frames works over MCP, and why it's the cheap path
Frames is an MCP server. Add it to Claude Code, Codex, Cursor or whichever agent your team already lives in, sign in once, and your agent gets nine Frames tools. Searching the catalog is free; only pulling data costs credits. Every request goes through three moves.
Search. Your agent describes what it needs, in up to four phrasings, and Frames returns the best matches from a catalog of about 37,000 paid endpoints, ranked by whether they actually delivered on past calls. Each match comes with a live check, a price and the exact inputs, so it's ready to call.
Invoke. Your agent calls up to ten tools in one request. Frames pays each provider in stablecoins on Base, Solana and Tempo, so nobody on the team holds an API key or a wallet, and you pay the cost of the data plus 15%, in credits. 1 credit is $0.001, and a typical call costs cents.
Check. Every response is checked for empty bodies, hollow results and errors dressed up as success. Calls that don't deliver are refunded, each request runs under your plan's cap (1,000 credits on the Free plan), and every purchase gets a receipt: the tool, the price, the transaction and whether it delivered.
Your own model does the thinking, so you pay for data and not for reasoning. That's what makes MCP the cheapest way to use Frames, and the natural one for a team that already works out of a coding agent.
For a hackathon team, that turns into a handful of prompts worth running in the first two days:
- Idea validation. Find out who built your idea before you, and what happened to them. "Use Frames to list every team that built a stablecoin payroll product on Solana or Base in the last 18 months: what they shipped, funding raised, whether the product is still live, and what users complain about on X and Reddit. Link every claim." Funding rounds and full complaint threads live in paid databases; a search engine surfaces the launch post and nothing after it.
- Market sizing for the pitch. Put a sourced number on the slide instead of an adjective. "Use Frames to size stablecoin payments on Solana for the last 90 days: transfer volume, active sending wallets, the top five apps by volume and month-over-month growth, with a source behind every number." The join across on-chain analytics and app-level data is an analyst product nobody publishes for free.
- Competitor teardown. Know the incumbent better than they know themselves. "Use Frames to build a brief on Jupiter: funding history, team size, web traffic trend over the last 12 months, fee structure, and the top complaints in reviews and forum threads." Traffic, headcount and sentiment data each sit behind their own paywall.
Live data in the demo. The same search and invoke calls are plain HTTP endpoints, so your backend can pull a live number straight from a provider with no MCP client in the loop: search once to find the tool, then invoke it with a cap. The demo runs on real 24-hour USDC volume instead of a mocked JSON file.
curl -X POST https://api.frames.ag/v1/tools/invoke \
-H "Authorization: Bearer $FRAMES_API_KEY" -H "Content-Type: application/json" \
-d '{
"calls": [{ "id": "<tool id from search>", "args": { "<input>": "<value>" } }],
"max_usd": 0.5,
"idempotency_key": "demo-usdc-volume-001"
}'The request and response shapes are on frames.ag/docs/tool-endpoints.
You never pick a provider by hand. Your agent describes what it needs, and Frames finds the source, pays for it and tells you whether the result arrived.
Frames inside Colosseum Copilot
Colosseum Copilot is a skill you install into your coding agent with one command. It's free with a Colosseum account and runs in Claude Code, Codex and OpenClaw.
npx skills add ColosseumOrg/colosseum-copilot -g
npx @colosseum-org/copilot-connect loginOnce it's in, your agent can search every Colosseum submission since Renaissance: 8,286 projects across five hackathons, each sorted into a category and tagged with the tools it used, plus 8,400+ live crypto products through The Grid and a research library that runs from Satoshi's emails to Paradigm and a16z crypto. Since v2 it also reads your own drafts and submissions, so it can review your entry against everything that came before it.
That archive answers what was built. For what's happening now, outside Colosseum, Copilot turns to Frames, over the same MCP connection your agent already uses.
- You ask Copilot a question
- Copilot answers from its own archive, free
- A gap remains: it proposes one capped Frames lookup and you approve
- Frames pays the tool and checks delivery, then Copilot reports the charge
The flow is deliberately conservative. Ask about an outside competitor, a funding round, current regulation, hiring or social reach, and Copilot does the free research first. If a gap remains that could change your decision, it reads your Frames balance with the free usage tool, then ends its answer with one offer: the gap, the single lookup that closes it, a cap (usually 250 to 500 credits, so 25 to 50 cents) and what share of your balance that is. Nothing is spent until you say yes. After the call, it reports the charge and your new balance.
It takes the cheap path on purpose: a free catalog search to find the one tool that answers, a free check of its price and inputs, then a single paid call. Your agent does the reasoning, so the credits go to the data and nothing else.
Two things are disclosed every time: Colosseum is an investor in Frames, and you pay with your own credits. A free grant is never treated as permission to spend.
Here's roughly how that reads after you ask whether anyone is already building invoice financing for Solana merchants (an illustration, not a transcript):
Copilot: three precedents from the archive, with the hackathon, the placement and what each one shipped, all of it free.
Then the offer: "Two of these teams may have raised since. Frames can check Messari's funding data for both in one lookup, capped at 500 credits (about $0.50, a sixth of your 3,000). Colosseum is an investor in Frames, and you pay with your own credits. Approve?"
The lookups go to sources like Nansen, for chain rankings and labeled wallets, and Messari, for funding rounds, investors and attention on a token. Copilot asks Frames for source URLs and as-of dates with every result, and keeps Frames' current evidence separate from its own historical records, so you always know which claim came from where.
It uses the Frames connection your agent already has: the Frames MCP server, or a FRAMES_API_KEY in your environment. In Claude Code, that's one line:
claude mcp add --transport http --scope user frames https://api.frames.ag/mcpSetup for Codex, Cursor and the rest is on frames.ag/docs/mcp.
Copilot tells you what 8,286 teams built before you. Frames tells you what's happening now.
The Crypto World's Fair: what's on the table
Colosseum's current hackathon, the Crypto World's Fair, runs from September 14th to October 12th, 2026. It's the first Colosseum hackathon open to every chain rather than Solana alone, and more than 9,000 builders have registered.
| Prize | Amount |
|---|---|
| Grand prize | $30,000 |
| Next 20 projects | $15,000 each ($300,000) |
| Public Good Prize | $5,000 |
| University Prize | $5,000 |
| Solana, Tempo, Hyperliquid and Zcash tracks | $100,000 each |
| Ethereum, Base, Arbitrum and Robinhood Chain tracks | $25,000 each |
That's $840,000 in prizes. Every submission competes in the general pool whatever chain it targets, and in its ecosystem track on top. The Solana track alone pays ten teams $10,000 each.
Winners are also considered for Colosseum's accelerator: $250,000 in pre-seed funding per accepted team, twelve weeks in San Francisco and a Demo Day. That's where the $2.5 million in seed funding comes from.
Submissions close on October 12th, 2026. Register at colosseum.com/worldsfair.
If you're entering, the Free plan's 3,000 credits are enough to validate the idea, source the pitch and put live data in the demo, and Copilot tells you what each call costs before it runs.
Getting started
Every account starts on the Free plan: 3,000 credits a month, no card. From there, Frames works wherever your team does.
- Sign up at frames.ag, then add Frames to Claude Code, Codex, Cursor, Claude or ChatGPT over MCP. Setup for each app is on frames.ag/docs/mcp.
- Inside Colosseum Copilot: install the skill and connect Frames as above. Copilot's own guide is at docs.colosseum.com/copilot.
- From your own app, through the tool endpoints: frames.ag/docs/tool-endpoints.
Plans start at $19 a month for 10,000 credits; the full table is on the pricing page.
Knowing the most wins
Hackathons are won by the team that knows the most about the problem, not the one that types fastest. Frames finds the data your idea needs, pays for it, checks it was delivered and hands you one receipt. Copilot tells you what every Colosseum team before you built. Between them, you get to the right idea faster, with numbers the judges can check.
If you're building for the World's Fair and something gets in the way, find us on Telegram.
Every account starts on the Free plan: 3,000 credits a month, no card, and access to the whole catalog. 1 credit is $0.001, so a typical lookup costs cents.
Inside Copilot, a direct lookup is capped at 250 to 500 credits, and Copilot asks before each one. If you run out, plans start at $19 a month for 10,000 credits.
Yes. Frames runs an MCP server at api.frames.ag/mcp, and any app that supports MCP can use it: Claude Code, Codex, Cursor, Claude, ChatGPT and more. Add it once, sign in, and your agent gets nine Frames tools; searching is free and only pulling data costs credits.
Copilot uses that same connection, or a FRAMES_API_KEY in your environment, so there is nothing extra to set up.
No. Frames is a research tool, and the data it reaches covers any ecosystem, so it is just as useful for a Base, Ethereum or Hyperliquid entry. Under the hood Frames pays providers on Base, Solana and Tempo, but that has nothing to do with the chain you build on.
Copilot's own archive is deepest for Solana and thinner elsewhere, which is exactly the kind of gap it offers to fill through Frames.
Every paid call is checked before the result reaches your agent: empty results, hollow answers and errors dressed up as success get caught, and calls that don't deliver are refunded. Every request also runs under a spending cap.
A receipt proves delivery, not truth. Ask your agent for the source behind every number, and check the ones that matter against the primary source before they go on a slide.
Marcelo Kunze
