What Builders Made with The Graph at ETHGlobal Lisbon 2026

The Graph put $15,000 in prizes across four categories at ETHGlobal Lisbon. Ten projects won, and every submission is public on the ETHGlobal showcase.

The tracks were not arbitrary. Each encoded a bet about where blockchain data is heading.

• The first was that the ecosystem needs tools more than it needs apps, so one track paid specifically for reusable infrastructure: MCP servers, agent SKILLs, plugins, payment tooling.
• The second was that agents are becoming a primary consumer of on-chain data, which changes what the data has to do; an interface can tolerate a stale number, an agent that acts on one cannot.
• The third was that standards compound, because a shared schema across every protocol of a type means one query pattern covers a whole category.

Here is who won. The pattern across them is at the end, and it is the part worth reading.

Best AI Tooling for The Graph · $5,000

This category includes projects that are built with reusable infrastructure that makes The Graph easier to build with from AI environments like Claude, Cursor, and ChatGPT. It was judged on how much it lowers the barrier for the next developer.

1. 🥇 Pista · $2,500. Smart contracts can check who is calling and when. They cannot see a wallet's history across protocols, or a slow drain only visible once you zoom out. Pista gives them a brain outside the EVM: a Substreams module streams every block into ClickHouse through a hosted sink, an AI agent scores it against adaptive behavioral baselines, and a flagged deviation triggers a response on-chain or off, pausing a contract, blacklisting an address, or raising an alert. It needs no integration from the protocols it watches, and flags deviation from a baseline rather than known signatures. Built by Mattia Careddu, Marco Crotta, Antonio Luca, Simone Barbanti, and Matteo Gagliardi.
2. 🥈 ArcBook · $1,500. An on-chain order book where every maker position is an executable pricing curve rather than a fixed-price level, so one position expresses an entire execution policy. A native Subgraph is the discovery layer for every curve, fill, and route, and the solver rejects stale snapshots before preparing calldata. Ryad Aouak also shipped a public Executable Liquidity MCP for AI clients.
3. 🥉 deeptrace · $1,000. Four read-only MCP tools on Messari Standardized Subgraphs, including a lending comparison that runs one query template against Aave v3, Seamless, and Moonwell. Responses are rejected unless the pinned deployment ID matches, so an agent gets provenance it can cite. Built by Matvii Nesterenko and Daniil Chernykh.

Best AI Use Case of The Graph · $3,000

Agents and applications using The Graph as a live source of blockchain data. Neither winner built a trading bot.

1. 🥇 ColdProof · $2,000. Cold-chain failures cost the biopharma industry an estimated tens of billions a year, and the evidence still often sits in editable data loggers controlled by one party. ColdProof puts a sensor box in the shipment that signs temperature and door events at the source, writes each one to an immutable consensus log, and settles escrow from that log: chain intact, the carrier is paid; breach, the shipper is refunded and the reason recorded. To make it queryable, Silvia Mogas and Adrian Tavella wrote a custom firehose-core provider that turns live Hedera Consensus Service messages into a Substreams-indexable block stream, so Graph Node can serve the sensor readings through GraphQL.
2. 🥈 BookerBob · $1,000. When an AI agent walks up to a business, the business can treat it as fraud and demand full prepay, or trust it and get farmed. BookerBob is a hotel-booking gateway where the answer to "who is behind this request" changes the risk terms and never the price. Underneath is context-bands-mcp, a standalone MCP server that reads an address through The Graph's decentralized network and returns four coarse bands: how far back the history goes, how often the address acts, how many venues it touches, what size it moves. Aave v3 and GMX are read through Messari standardized schemas, adding a source is a registry entry rather than code, and raw figures never leave the server. Built by Max Choly and Dan Surfer.

Best AI Use Case of The Graph (Continuity) · $4,000

A separate pool for projects carried forward from earlier work rather than started on Friday night.

1. 🥇 EQLTY · $2,000. State one investment objective in plain language, set a budget, and four bounded agents take it from there: Scout finds candidates, Risk checks policy and freshness, Trader prepares a guarded trade, Auditor reconciles the result against indexed evidence. A parameterized Substreams package filters Uniswap V4 stock token pool events, binding 94 pools to their tickers without redeploying, and that evidence sits inside the agents' prompts. The Risk gate only advances a trade when the evidence carries live provenance, a canonical transaction hash, and acceptable freshness; stale data blocks the candidate rather than quietly weakening the run. Built by Julio Cruz.
2. 🥈 Justify · $1,000. A social prediction market on Base where creators launch markets and earn a share of the fees. The Gnosis fixed-product market maker emits no price, so Vitalii Basiuk and Vadym Kotov wrote mappings that rebuild pool balances from event deltas with zero RPC calls, computing PnL, price series, and the full resolution lifecycle inside the index.
3. 🥉 Binary Stamp · $1,000. Hash a file in the browser, register the hash, and hold a timestamped, independently verifiable record of authorship. Most disputes turn on sequence, who had it first and whether the artefact changed since, and a stamp answers both for cents at creation rather than at conflict. The Graph indexes stamps and transfers, which makes the registry instantly searchable. Built by Antonio Roldao.

Best Use of Composable or Standardized Graph Products · $3,000

Write the query once, run it against every protocol that conforms.

1. 🥇 Am I cooked · $2,000. A wallet scan with a fun half and a serious half. The fun half surfaces money you forgot, in pools everyone abandoned. The serious half traces every live approval, re-verifies it against the chain, matches spenders against a registry of documented hacks, and returns the exact revoke transaction, with no wallet connect and no signature. The Graph is the evidence layer: Messari standardized schemas mean one query format covers Aave v3, Compound v3, and Spark, and adding a conforming market is one registry line. "One query, N protocols," in the team's words, replaced a week of writing adapters. That is not merely convenient, because a checker covering two lending protocols tells you that you are safe while the third one drains you. Built by Viacheslav Smirnov and Nikita Shliakov.
2. 🥈 atlas · $1,000. Ask a question about on-chain activity and get back a live mini app: an interface generated from real Graph data that watches, and, if permitted, acts inside a policy the user sets. The architectural move is that the resolver picks a schema family, lending-cdp or dex-amm-extended or perp-futures, rather than a single Subgraph ID, so one question fires the same query shape at every healthy deployment of that family and merges the results. A question nobody anticipated still resolves, because the schema is shared even when the protocol is new. Fabian Ferno composed Standardized Subgraphs across 86 verified deployment IDs, a Substreams gRPC subscription driving per-block triggers, and x402 payment, and open-sourced an MCP server and a SKILL.md.

What the Winners Had in Common

Ten teams, working independently under time pressure, converged on three things nobody asked them to do:

1. Standardized schemas were the biggest force multiplier, and every team that used them built the same missing piece. Am I cooked, atlas, BookerBob, and deeptrace each arrived at the same architecture: write one query against a shared schema, then register new protocols as data rather than code. All four hand-built a registry mapping schemas to deployments, which signals both that standards work and that the index of what conforms to them is missing.
2. Provenance became critical rather than decorative. deeptrace rejects any response whose deployment ID does not match a pinned value. EQLTY blocks a trade when freshness or block lag fails. BookerBob returns “unavailable” rather than a confident wrong answer. atlas health-checks its sources before it spends. When an agent’s output is a transaction instead of a chart, “probably current” stops being an acceptable input and _meta stops being a footnote.
3. Real-time streaming showed up as a correctness requirement, not just a performance attribute. atlas moved its triggers off Subgraph polling onto a Substreams gRPC subscription, with cursor resumption and reorg handling, because a guard five minutes late is a preference for an app that displays and a bug for one that spends. Pista streams every block into ClickHouse to score behavior against a baseline, and ColdProof wrote a firehose-core provider to bring a new data source into the stack entirely.

One more thing is worth naming: the winners were not all DeFi. Pharmaceutical cold-chain custody, file provenance, hotel underwriting. The through line is not finance, it is that each needs on-chain history that is structured, live, and honest about how fresh it is.

Build on The Graph

Thanks to ETHGlobal, and to every team that built with The Graph in Lisbon, including the many not listed above.

Every project is on the ETHGlobal showcase, and most winners are open source, which makes several of them a faster starting point than a blank repo. If the tooling projects caught your attention, the AI Suite documentation covers the Subgraph MCP. If the standards work did, start with Standardized Subgraphs. Get an API key in Subgraph Studio and browse Graph Explorer.

If you build something on top of one of these, the Discord is where to say so.

About The Graph

The Graph is a suite of blockchain data infrastructure products that extract, process, and deliver scalable blockchain data solutions across 60+ networks. The Graph enables application developers, data analysts, AI agents, and enterprise teams that need structured, real-time access to blockchain data. Products include Subgraphs, Firehose, Substreams, and Amp. As of early 2026, The Graph has served over 1.27 trillion queries to more than 75,000 projects, powered by a network of independent Indexers around the world.

Follow The Graph on X, LinkedIn, Instagram, and Reddit. Join the community on The Graph’s Telegram, join technical discussions on The Graph’s Discord.