Querying Blockchain Data with Natural Language: MCP and Skills from The Graph
As AI tools become a standard part of the developer workflow, the way users access blockchain data is changing. integrations and AI agent skills are creating new interfaces for interacting with onchain data, allowing both developers and non-technical users to query live protocol data through natural language rather than writing GraphQL or SQL by hand.
The Graph ecosystem has been building toward this shift. With MCP servers now available for Subgraphs and a growing library of agent skills for and , AI assistants can connect directly to The Graph’s data infrastructure. This post covers what’s available today, how it works, and what’s coming next.
Background: What is a MCP?
Model Context Protocol (MCP) is an open standard, originally developed by , that enables AI assistants and AI agents to connect to external data sources and tools. Rather than relying solely on training data or user-provided context, an MCP-enabled assistant can reach out to live data in real time, querying APIs, reading schemas, and returning structured results within a conversation.
For blockchain data, MCP addresses a longstanding access problem. Onchain data is open and transparent by design, but querying it has traditionally required specialized knowledge: writing GraphQL queries, understanding Subgraph schemas, or running custom indexing infrastructure. MCP collapses that workflow by letting AI handle the query construction and execution on the user’s behalf.
MCP Integrations Available Today
Subgraph Search MCP
The provides access to over 15,000 publicly available across The Graph Network. Through a standardized interface, AI assistants can search for relevant Subgraphs by keyword or contract address, inspect GraphQL schemas, and execute queries against specific deployments. Compatible clients include Claude, Cursor, and Cline.
Because The Graph hosts a large volume of public Subgraphs, covering major protocols like Uniswap, Aave, Compound, ENS, and many others across Ethereum, Arbitrum, Base, Polygon, Optimism, BSC, and additional chains, the Subgraph MCP enables broad, cross-protocol queries without requiring users to deploy or maintain their own data infrastructure.
Substreams Search MCP
The lets AI agents search, inspect, and analyze packages through natural language. This covers the full path from registry discovery to sink deployment. Through four tools, an assistant can search the package registry by keyword and filter by network, inspect any package's full module graph, protobuf types, and dependency DAG, list a package's modules in lightweight form, and analyze a package's sink configuration to generate ready-to-run CLI commands for deployment. In practice, this means a developer can ask for "a Uniswap package on Polygon," see exactly what data it produces, and get the SQL schema and commands needed to start sinking that data—all without manually browsing the registry or decoding .spkg files by hand. The server runs with no installation via npx and supports both local clients like Claude Desktop, Claude Code, and Cursor and remote agents such as OpenClaw (SSE/HTTP), making it usable across a wide range of AI development setups.
Use Case Specific MCPs
In addition to the core MCP servers, community developers have built more targeted integrations. Notable examples include the , which exposes 14 tools for querying Aave V2/V3 lending markets, governance data, user positions, and liquidation events; and the , which provides unified access to 40+ lending protocols across multiple chains using Messari’s standardized Subgraph schemas.
The Lending MCP, in particular, demonstrates the value of standardized data. Because the underlying Subgraphs share a uniform schema, a single natural-language query, such as “where can I find the best stablecoin lending rates right now?”, can fan out across dozens of protocols and return comparable, structured results. Early testing has shown that the tool not only retrieves rate data but can contextualize it, explaining why certain strategies may be more effective than others based on current market conditions.
From MCPs to Skills: A New Paradigm for AI Agent Development
MCP servers provide AI assistants with real-time data access, but each one requires manual configuration, including editing a config file, adding an API key, and restarting the client. As the number of available MCPs grows, this setup process can become a barrier, particularly for non-technical users.
Skills represent the next layer of abstraction. A skill is a downloadable package that bundles expert knowledge, tool configurations, and MCP access together so that an AI assistant gains a complete capability set without manual setup. Instead of configuring individual MCP connections, users install a skill, and the assistant has everything it needs to work with a given product or domain.
Subgraph Skills
Skills are now available to aid in Subgraph development. The provides a collection of open-source skills that give AI workflows expert-level knowledge for building, testing, and optimizing Subgraphs. Developers can describe what they want in natural language instead of memorizing schema syntax and manifest configuration. Available as a Claude Code plugin (and in an OpenClaw format for other agent frameworks), the package currently includes three skills:
Subgraph Development (subgraph-dev)
Core development knowledge spanning schema design and GraphQL types, manifest configuration (subgraph.yaml), AssemblyScript mapping handlers, data source templates, and contract bindings. It also covers Subgraph Composition for combining multiple Subgraphs, Subgraph Uncrashable for safe code generation, and ready-made patterns for common contract types like ERC-20, DEXs, NFTs, lending, staking, and governance protocols, so an assistant can scaffold a working Subgraph for a given protocol from a single prompt.
Subgraph Optimization (subgraph-optimization)
Performance best practices drawn directly from The Graph's documentation, including pruning with indexerHints, using @derivedFrom for arrays, immutable entities, Bytes as IDs, avoiding eth_calls, timeseries and aggregations, and grafting for hotfixes. With this skill installed, an assistant can review an existing Subgraph and recommend concrete changes to speed up indexing and reduce query latency.
Subgraph Testing (subgraph-testing)
A complete quality-assurance toolkit built around Matchstick and the Subgraph Linter. It covers static analysis to catch bugs before runtime, unit testing setup and patterns, mocking events and contract calls, entity assertions, data source mocking, and CI/CD integration, along with a troubleshooting guide for common indexing errors.
Together, these skills lower the barrier to Subgraph development considerably. Rather than reading through documentation across schema design, mapping logic, optimization, and testing, a developer can install one package and have an AI assistant that already knows The Graph's best practices. This means turning prompts like "create a schema for tracking DEX swaps" or "write unit tests for my Transfer handler" into working, optimized code.
Substreams Skills
This pattern is also in production for Substreams. The , developed by StreamingFast, provides a collection of open-source skills that give AI assistants expert-level knowledge for Substreams development. Available as a Claude Code plugin, the package currently includes:
Substreams Development (substreams-dev)
Comprehensive guidance on creating substreams.yaml manifests, writing efficient Rust modules (map, store, and index types), designing protobuf schemas, performance optimization, and debugging common issues.
Substreams SQL (substreams-sql)
Expert knowledge for building SQL database sinks, covering both Database Changes (CDC) and Relational Mappings approaches, with patterns for PostgreSQL and ClickHouse, including analytics-optimized schemas and time-series patterns.
Substreams Testing (substreams-testing)
A complete testing strategy covering unit testing with real blockchain data, integration testing, performance benchmarking, and CI/CD pipeline integration.
Installation is straightforward. In Claude Code, users run a single command to add the plugin and install the desired skills. The skills are also compatible with Cursor and VS Code (1.107+), with installation requiring only a directory path in the IDE’s settings.
The vision across The Graph ecosystem is to extend this skills model to each Subgraphs and Substreams so that each has a corresponding skill package giving AI agents access to the relevant MCPs, documentation, and domain expertise in a single installation.
Practical Applications
The combination of MCP servers and skills opens a range of use cases across The Graph data products:
- DeFi analysis: Query lending rates, liquidity pool depths, swap volumes, and yield opportunities across protocols and chains through natural language. , for example, can compare stablecoin rates across 40+ protocols in a single query.
- Substreams development: With the Substreams skills installed, AI assistants can guide developers through building, testing, and deploying high-performance indexing pipelines—from manifest configuration through production optimization.
- Research and content: Writers and analysts can access real-time onchain data without learning GraphQL, enabling data-driven reporting on protocol activity, governance trends, and market dynamics.
- AI agent development: Autonomous agents can use The Graph MCP servers to access blockchain data programmatically, supporting use cases from automated trading analysis to onchain monitoring and alerting.
Getting Started
Detailed documentation is available for each MCP and Skills repo:
Subgraph MCP: thegraph.com/docs/en/ai-suite/subgraph-mcp/introduction/
Subgraph Skills: https://thegraph.com/docs/en/subgraphs/skills/
Graph Lending MCP: github.com/PaulieB14/graph-lending-mcp
Graph Aave MCP: github.com/PaulieB14/graph-aave-mcp
What’s Ahead?
MCP and skills represent the first phase of The Graph’s broader AI integration strategy. As outlined in the 2026 technical roadmap, upcoming developments include A2A (Agent-to-Agent) integrations that enable AI agents to communicate with each other through The Graph data layer, and x402 payment support that allows AI agents to autonomously query the network and pay per-query without requiring pre-configured API keys.
As additional skills packages are developed for Subgraphs, Substreams, and Amp, the goal is a consistent experience: install a skill, and the AI assistant gains full access to the relevant data product’s capabilities. Combined with The Graph’s open data infrastructure—over 15,000 public Subgraphs, pre-indexed token data across 10+ chains, and high-performance streaming via Substreams—these tools are making blockchain data accessible to a broader range of users and applications.
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 , , , and . Join the community on The Graph’s , join technical discussions on The Graph’s .
