The Graph

The Graph launched a set of open-source Substreams Skills that let AI coding assistants (Claude Code, Cursor, and others) turn a plain-language prompt into a working blockchain data pipeline in minutes, instead of days of hand-written Rust.

• Skills cover EVM ABI decoding, Solana account models, Bitcoin UTXO, SQL sink selection, testing, deployment, and the Portal API, including dev, ethereum, solana, bitcoin, convert-sql, sink, sink-deploy-local, hosted-sink, and testing.
• A dedicated substreams-convert skill maps an existing Subgraph's schema.graphql entities to protobuf types and its AssemblyScript handlers to map and store modules, carrying initialBlock over — so a Subgraph hitting its limits becomes a stream that can be sunk anywhere.
• A companion machine-readable OpenAPI spec for the Substreams registry lets agents find packages without scraping HTML.

The Graph introduced the Subgraph MCP (Model Context Protocol), which lets AI agents drill down into onchain data. A sequence of four calls takes an agent from a protocol name to a live query: search subgraphs by keyword, get deployment 30D query counts, get schema, and execute query.

The Graph is joining ETHGlobal's ETHOnline hackathon, offering $15,000 across two tracks: Best AI Project on The Graph and Best Use of Composable or Standardized Graph Products.

• $15,000 total prize pool split across two tracks.
• Entries must use live data; mocked or static datasets are disqualified.
• Runs during the ETHOnline hackathon.

The Graph (via StreamingFast) introduced Hosted Stores, a managed, block-aware key/value store for Substreams modules. It lets a module look data up mid-stream without recomputing it on every block, and is reachable over gRPC or as a store input.

Hosted Stores removes the need to maintain a self-hosted store sidecar when a Substreams module needs on-chain lookups while executing, cutting infrastructure overhead for pipelines that previously had to fetch or re-derive state across blocks.

The Graph Foundation published Part 2 of its technical roadmap series, officially announcing the Foundation's new mandate and how it will bring the roadmap to life. Part 1, published earlier this year, laid out the technical future of The Graph; this second installment focuses on the Foundation's role in executing it.

The Graph's Agent0 Subgraphs now index all on-chain activity generated by ERC-8004, which is live on five networks. Every agent registration, metadata update, and feedback entry emits on-chain events that Agent0 captures under a single GraphQL schema.

This lets an agent find and score a counterparty in a single query, instead of stitching together data across chains and formats.

SWAN, a protocol building autonomous wealth management on Robinhood Chain, is using The Graph's Substreams to stream vault activity in real time. The data powers SWAN's vault registry, activity feed, and live TVL, letting its agent respond to market changes and execute onchain rebalances within each vault's strategy and risk parameters.

The Graph released Substreams v1.20.0, bringing end-user improvements to indexing reliability and performance.

• Backfills finish faster and hold less data in memory.
• Reorgs no longer leave bad data behind on deletes.
• Failures developers already fixed stop replaying from cache.
• Errors now surface instead of being silently handled.

Crypto exchange MEXC now supports GRT on the Arbitrum network. GRT is The Graph Network's native work utility token, and the protocol's contracts have run on Arbitrum since the network migrated there.

With the listing, users can delegate in The Graph network directly on Arbitrum, making the process notably simpler.

The Graph updated its data platform at thegraph.com to organize blockchain data collections by relevant categories, making it easier to find the specific data a user needs.

Categories target use cases like testing strategies, giving governance updates, and tracking funds across bridges and marketplaces, replacing a search experience organized around who published the data with one based on what the data is.