Dune Launches “Dune Agent,” an AI‑Powered Data Discovery Chatbot for On‑Chain Analysts
By [Your Name] – March 3, 2026
Dune, the leading analytics platform for blockchain data, announced the beta release of Dune Agent, a conversational assistant built on Anthropic’s Model Context Protocol (MCP). The tool is designed to streamline the search for tables, decoded events, and curated “Spells” amid Dune’s ever‑growing catalog of on‑chain datasets.
The data‑discovery dilemma
Dune’s marketplace now hosts millions of tables spanning dozens of blockchains. New chains are added regularly, protocols are decoded, and community contributors continuously publish Spells—pre‑built query pipelines that turn raw blockchain logs into ready‑to‑use datasets. While this breadth makes Dune a one‑stop shop for DeFi analytics, it also creates a bottleneck: analysts spend disproportionate time locating the right tables, tracking schema changes, and identifying which Spells could accelerate their investigations.
The issue is not limited to individual users; protocol teams, data engineers, and researchers all face the same friction when trying to bridge the gap between “I want to analyze X” and a functional SQL query.
How Dune Agent tackles the problem
Instead of relying on conventional large language models (LLMs) that operate within static context windows, Dune Agent leverages MCP to give the model a live, dynamic view of Dune’s internal search infrastructure. In practice, the chatbot can query the platform’s up‑to‑date catalog in real time, returning exact table names, contract addresses, and Spell definitions without the hallucinations that plague typical LLM‑only approaches.
Key components of the solution include:
| Feature | Description |
|---|---|
| MCP‑backed search layer | A set of servers that act as an interface between Anthropic’s model and Dune’s native search APIs. |
| Specialized lookup tools | Separate endpoints for raw tables, decoded events/calls, and curated Spells, each supporting fine‑grained filtering by blockchain, schema, and contract address. |
| SQL extraction for Spells | The agent can surface the underlying transformation logic of a Spell, allowing analysts to adapt or extend existing pipelines. |
| Contextual relevance scoring | Results are ranked for speed and relevance, helping users find the most pertinent data quickly. |
The beta is currently at version 0.1, and Dune is inviting users to test the interface at dune.com/agent and share feedback through dune.com/feedback.
Early impact and user scenarios
- Analysts: The chatbot surfaces previously unknown Spells that match a user’s investigative intent, cutting down query‑building time from hours to minutes.
- Protocol teams: Developers can instantly retrieve every table linked to their smart contracts across multiple chains, facilitating audits and upgrades.
- Data engineers: By exposing dependencies between raw logs and curated datasets, the tool reduces the need to comb through GitHub repositories for lineage information.
Initial internal testing suggests noticeable improvements in search speed and answer accuracy compared with prior manual or static‑LLM methods.
Why MCP matters for blockchain analytics
Traditional LLMs are constrained by a fixed token window, forcing developers to pre‑load a snapshot of searchable content. In a domain that evolves daily—new token contracts, emergent Layer‑2 solutions, and community‑generated Spells—such a static approach quickly becomes obsolete. MCP sidesteps this limitation by allowing the language model to call external APIs on the fly, effectively turning the AI into a “data‑aware” assistant rather than a static knowledge base. For Dune, this means the agent can always reflect the most recent additions to the platform’s data catalog.
Key takeaways
- Dynamic AI‑augmented search: Dune Agent demonstrates that coupling LLMs with real‑time query capabilities can overcome the scalability challenges of massive blockchain data warehouses.
- Reduced friction for on‑chain research: By automating table discovery and Spell identification, the tool promises to shave hours off analysts’ workflow.
- Broad utility across roles: The assistant is positioned to serve analysts, protocol developers, and data engineers alike, enhancing cross‑functional collaboration.
- Early stage, community‑driven: At version 0.1, Dune is actively seeking feedback, indicating a roadmap that will likely incorporate additional filters, richer explanations, and tighter integration with Dune’s SQL editor.
- MCP as a blueprint: The implementation may inspire other DeFi infrastructure providers to adopt model‑driven context protocols for their own data discovery challenges.
Outlook
If the beta feedback is positive, Dune could soon roll out a full‑featured version that integrates directly with its query editor, supports multi‑modal inputs (e.g., natural‑language to SQL translation), and possibly opens the MCP layer to third‑party data providers. Such enhancements would reinforce Dune’s position as the de facto analytics hub for the decentralized finance ecosystem, where rapid, accurate data retrieval is increasingly critical for competitive edge.
For a hands‑on preview, visit dune.com/agent and share your experience via dune.com/feedback.
Source: https://dune.com/blog/dune-agent
