We propose the development of the Cardano Governance MCP: natural language access to governance data—an open-source MCP Server designed to transform fragmented governance data into structured, semantic, and machine-readable context.

As Cardano governance expands through the Intersect Constitution, ICC resolutions, DRep voting, and Catalyst community proposals, participants must navigate increasingly complex and unstructured information scattered across on-chain data, GitHub repositories, Intersect communications, and Catalyst portals. This proposal addresses that fragmentation by delivering a tool to unify and contextualize governance data. The MCP Server will allow voters, DReps, vendors, and governance institutions to access clear, queryable, and traceable records of governance behavior and decisions.

Use cases supported by the MCP Server:

Ask about DReps details like ID, status, voting power, objectives, motivations, qualifications, and more.

E.g.: What are the main motivations for XXXX DRep?

E.g.: Which DReps align with climate-related objectives?

Ask about current proposals, both off and on-chain.

E.g.: Is there any on-chain proposal live right now?

E.g.: Is there any AI proposal being voted on GovTool right now?

E.g.: Are these proposal objectives overlapping with another active proposal?

Since we are using LLMs, questions can be formulated in any language.

E.g.: el DRep XXXX está activo actualmente?

E.g.: この提案の目的は、他のアクティブな提案と重複していますか？

Combine all use cases.

E.g.: Who is the DRep with the highest voting power that voted in the ZK Bridge proposal?

E.g.: What high voting power DReps haven’t voted for the ZK Bridge proposal yet?

E.g.: I greatly support Africa-related proposals. To which DRep can I delegate my ADAs? The DRep should have voted YES on Africa-related proposals last year.

Future developments could include enabling direct interaction with the Cardano blockchain, allowing users to delegate ADA to their chosen DRep simply by conversing with the LLM. Additional enhancements may involve integrating more data providers such as Blockfrost to enrich the context and improve the accuracy of responses.

Cardano's governance ecosystem is growing rapidly, with diverse actors such as voters, MBOs, Founding Entities, DReps, vendors, and individuals actively engaging with tools like Cardano GovTool. However, despite the growing functionality of GovTool, it remains challenging for these users to extract meaningful insights or access tailored information about governance activities.

Key limitations include:

Poor accessibility for non-technical users: Normal voters struggle to query or filter data that matters to them (e.g., "Which DReps tend to support proposals related to Africa?").

Vendors and governance actors face data overload: There is no easy way to search, summarize, or interact with relevant governance data using natural language.

Fragmented sources and lack of semantic structure: Information is distributed between the GovTool backend, on-chain governance data, and external APIs like Koios, making integration difficult for AI or dashboard systems.

Without accessible, context-rich data interfaces, key stakeholders risk disengagement, and the broader ecosystem cannot benefit from LLM-based interfaces that could enhance understanding and participation.

Proposal Overview

We propose the development of a tool that provides LLMs with information from GovTool and other sources (like the blockchain itself), powered by the Model Context Protocol (MCP). This tool, an MCP Server, will allow voters, DReps, vendors, and governance institutions to:

Ask natural language questions about governance (e.g., "Which DReps align with climate-related proposals?", "Are these proposal objectives overlapping with another active proposal?").

Receive human-readable summaries of governance activity.

Easily filter and extract structured information from the GovTool backend and blockchain data sources.

The core of this system is a Model Context Protocol (MCP) Server implementation that connects with a governance context layer to the GovTool backend and Cardano data providers (e.g., Koios), transforming fragmented data into AI-ready structured formats.

Increased accessibility: Normal users can interact with governance data through simple questions.

Informed decision-making: DReps, vendors, and voters can better understand patterns, delegations, and proposal impacts.

Governance transparency: Enhanced visibility into DRep behavior, proposal evolution, and vote records.

Developer extensibility: LLM-ready MCP format enables integration into dashboards, bots, and civic tools.

1. MCP Server code

• Open-source repository in GitHub including code and documentation
• Package release in npm/pypi

2. Demos with MCP Hosts and LLMs

• Using various MCP Hosts and LLMs with multiple queries for each kind of user. E.g.: Claude Desktop with Claude 3.5 Haiku or 3.7 Sonnet, VS Code Agent Mode with a local LLM, VS Code and Cline or Roo Code, etc.

3. Developer Documentation

• Prompt benchmarks with multiple LLMs (trying to provide information about how good does the LLM X or Y respond to a specific prompt).
• Prompt templates and example use cases.
• Architecture diagrams (e.g.: governance context layer using GovTool backend, Koios, etc, connecting to the LLM).
• Usage tutorials

AI Systems Architect (1 FTE x 3.5 months) $39,375 (₳ 78,750)

Full Stack Engineer (1 FTE x 3.5 months) $28,440 (₳ 56,880)

Project Manager / Coordinator (1 PMs x 3.5 months) $8,750 (₳ 17,500)

Total $76,565 (₳ 153,130)

Estimated Duration: 3.5 months

Team Composition:

• 1 LLM/AI Systems Architect
• 1 Full Stack / API Developer
• 1 Project Manager / Coordinator: Delivery, documentation and community engagement

Eryx is a worker-owned labor cooperative with almost 15 years of experience solving complex problems involving mathematics and software. We are a group of nerdy PhDs with a solid background in math and computer science. Our team has extensive experience in blockchain development, cryptography, and privacy-focused protocols, and also in AI projects involving LLMs, Computer Vision, Data Science and Operations Research which gives us the expertise required to successfully deliver the project. Among other things, we’ve been core contributors to the Lambdaworks library and implemented the CircleSTARKS protocol in GPU alongside with Nethermind. This involved implementing several ZK cryptographic primitives (field operations, polynomial commitment schemes, FRI protocols, FFTs), in different platforms and languages. Furthermore, we implemented a proving system based on the Plonk protocol. We also developed a Plonky2 backend for Aztec's Noir, funded by the Ethereum Foundation. In connection with this proposal, we bring proven experience delivering large-scale AI projects for major Argentinian and multinational clients—including Adecoagro, Tenaris, and Galicia Seguros—spanning LLMs, computer vision, and high-volume data processing.

The MCP Server specification, reference implementation, and tooling will be hosted on a public GitHub repository under an MIT license. This repository will include:

• Version-controlled releases.
• Use examples and documentation.
• CI/CD pipelines to ensure stability.

Sustainability and funding

To ensure ongoing development, we plan to:

Apply for future Catalyst funding (e.g. Governance Support or Developer Ecosystem categories).

Collaborate with Intersect and governance-related DAOs to support maintenance.

Explore co-funding and partnerships with dashboard builders, data providers, and LLM infrastructure projects.

This tool will remain governance-neutral, extensible with more information providers and features—intended for broad adoption across the Cardano ecosystem.

The need for contextualized, accessible, and machine-readable governance data has been raised repeatedly across Cardano’s governance conversations, technical working groups, and community tools. Multiple signals highlight the strong alignment between the MCP Server and current ecosystem demands.

Since we are developing an MCP Server, we do not depend on a specific LLM but rather on the user’s preferred LLM and MCP Host (e.g.: such as Claude Desktop, VSCode, Cline or even MCP SDKs like modelcontextprotocol/python-sdk standalone or integrated into Pydantic, LangChain or AutoGen frameworks). Results may vary depending on the chosen LLM (we will provide a preferred LLM for the dev team).

We depend on the GovTool backend and the Koios API to gather the information.

This proposal involves only client-side code querying those dependencies, therefore there will be no need to build and maintain infrastructure like cloud servers and storage (thus, reducing costs).

There will be an open-source GitHub repo with the MCP Server code. There won’t be any fees or subscriptions from the open-source software.