The Problem: Data Discovery at Scale

Dune hosts millions of tables across dozens of blockchains. We are constantly adding new chains, decoding new protocols, and community members are contributing new Spells (curated datasets). Even as the former lead of the Spellbook project, I struggled to keep track of what new analytical capabilities were being unlocked.

This isn't just my problem—it's everyone's problem. Analysts waste hours searching for the right tables, understanding schema changes, and figuring out which Spells might accelerate their analysis. The friction between "I want to analyze X" and actually querying the data is too high.

Our Solution: MCP + Intelligent Search

We built a data discovery chatbot powered by Anthropic's Model Context Protocol (MCP) that fundamentally changes how users explore Dune's data catalog. Here's why MCP was the key:

Traditional approach: LLMs with static context windows struggle with dynamic, massive datasets. They hallucinate table names, miss recent additions, and can't effectively search millions of entries.

MCP approach: Instead of stuffing everything into context, MCP lets our chatbot dynamically query Dune's existing search infrastructure. It's like giving the LLM a live connection to our database catalog rather than a outdated snapshot.

The technical implementation:

• Built MCP servers that wrap Dune's internal search APIs
• Created specialized tools for searching raw tables, decoded events/calls, and Spells
• Implemented smart filtering by blockchain, schema, and contract address
• Added SQL retrieval for Spells to expose transformation logic

The result? Ask "Can I find all DEX trades on Arbitrum" and get not just table names, but understand which decoded contracts are available, what Spells aggregate this data, and even see the underlying SQL logic.

Real Impact for Real Users

This isn't just theoretical. In testing, we've seen:

• Analysts discover relevant Spells they didn't know existed, cutting query development time
• Protocol teams instantly find all tables related to their contracts across every chain
• Data engineers understand dependencies between raw and curated datasets without diving into GitHub

The chatbot doesn't just list tables—it explains relationships, suggests optimization paths, and reveals the "why" behind our data architecture.

Try It Now

The MCP data discovery chatbot is live in beta for users with paid accounts at dune.com/agent. We're specifically looking for feedback on:

• Search relevance and speed
• Explanation clarity for complex data relationships
• Missing features that would accelerate your workflow

This is version 0.1. Help us build the data discovery experience that on-chain analysts deserve.

Check out it at dune.com/agent

Give us your feedback at dune.com/feedback