MCP-1st-Birthday/agentic-economics
An entry by the Covia team for the Hugging Face 1st Birthday hackathon.
Team:
Also available on GitHub: https://github.com/covia-ai/agentic-economics-hackathon
Vision
Agents shouldn’t just answer questions – they should coordinate, contract, and transact.
Our goal is simple: show a glimpse of what agentic economies looks like when you move beyond glue code and brittle workflows, and treat agents as economic actors operating on top of a shared grid.
This project is our exploration of what happens when you give AI agents:
- a shared orchestration grid,
- a real system of record,
- and the ability to form and execute economic contracts.
For the MCP 1st Birthday Hackathon – Track 1, we use the Covia Federated Orchestration Grid together with MCP, Anthropic Claude and Cursor to let agents:
- negotiate terms,
- draft and validate economic contracts,
- and execute them end-to-end across tools and providers.
Demo
Covia Team Hugging Face demo link here:
- link to YouTube demo - https://youtu.be/7KkT8efTv1s
- link to HuggingFace space - https://huggingface.co/spaces/MCP-1st-Birthday/agentic-economics
- Social Media post - https://x.com/mikera/status/1995277731609653316
Agentic Economic Contracting with Covia.ai (MCP Hackathon – Track 1)
Agents shouldn’t just answer questions – they should coordinate, contract, and transact.
This project explores what becomes possible when you give AI agents:
- a shared orchestration grid,
- verified, governed tool access via MCP,
- multi-provider cognitive capabilities (Claude, OpenAI, Gemini, Cursor),
- a powerful DLT for economic assets and smart contracts (Convex),
- and a real, append-only system of record
For the MCP 1st Birthday Hackathon (Track 1), we built a prototype where agents negotiate, draft, validate, and execute economic contracts — end-to-end — across Anthropic, OpenAI, Manus, and the Covia Grid.
This is our first glimpse of agentic economics: agents acting as economic participants, not chatbots.
Project Overview
This project demonstrates an agentic economic coordination system built using:
- MCP (Model Context Protocol) for unified tool/model interoperability
- Anthropic Claude for negotiation + reasoning
- OpenAI tools for structured execution
- Google Gemini (optional) for multimodal analysis
- Covia.ai Federated Orchestration Grid for shared memory, contract state machines, and a governed system of record
What the agents actually do
Agents:
- Propose economic terms
- Negotiate until agreement
- Draft a structured JSON contract
- Validate through Claude/Cusror + MCP
- Execute actions using Claude tools
- Commit all state to the Covia Grid
- Produce an auditable, append-only ledger
This is a proof-of-concept for multi-agent, multi-LLM economic coordination.
Problem Statement: Why We Built This
Today’s agent frameworks break when you attempt real multi-agent coordination:
- no shared memory
- no shared system of record
- no contract/state model
- no cross-LLM cooperation
- brittle tool chains
- hidden execution context
If agents are ever going to participate in commerce, operations, logistics, payments, or negotiations, they need:
- verifiable contracts
- cross-provider cognition
- tool-level governance
- state synchronisation
- append-only auditability
This project demonstrates the earliest version of that future: agents as economic actors, not isolated chatbots.
What’s Unique About Our Approach
Most agent demos rely on:
- prompt wrappers
- LangGraph spaghetti
- static workflows
- single-LLM dependence
- hidden state
Our approach is different:
- Multi-provider cognition (Claude, ChatGPT, Cursor, Manus, Gemini etc.)
- MCP as a flexible protocol for accessing economic systems
- Covia Grid as a shared execution + memory layer
- Full end-to-end system of record
- Economic contract objects (not JSON hacks)
This combination enables negotiation, validation, execution, and reconciliation across agents — something no existing framework provides.
How We Used MCP (Track 1 Requirements)
This submission is built specifically for Track 1: Build with MCP.
We use MCP in three ways:
A. MCP as the universal transport layer
MCP lets Claude, OpenAI, Gemini, and Manus communicate through:
- a shared message schema
- unified tool calling
- model-agnostic interoperability
- consistent tool validation responses
This removes 90% of typical agent-framework brittleness.
B. MCP + Claude for safe tool execution
Claude is our MCP server providing:
- contract validation tools
- economic rule enforcement
- secure sandboxed execution
- deterministic error handling
All tool calls go through MCP → Claude/Cursor → Covia for recording.
C. MCP as the foundation of agent-to-agent coordination
Agents provide tool outputs to each other via MCP-triggered events, including:
- contract drafts
- negotiation summaries
- validation results
- execution confirmations
MCP ensures ensure all in interactions have a consistent, verifiable structure and are usable for a wide range of agentic tools.
Usage
Helpful tools and links
The space running the MCP server is:
https://mcp-1st-birthday-agentic-economics.hf.spaceYou can explore the Covia Grid, look at job history etc. with the Covia App - connect to the venue at https://mcp-1st-birthday-agentic-economics.hf.space to browse jobs, assets and operations.
Access to the Convex test network used in the demo is also available via a Hugging Face Space which provides an explorer interface, REST API endpoints and also has its own MCP tools:
https://mikera1337-convex-testnet.hf.spaceComponents Included in This Submission
This repo includes:
Dockerfile→ Dockerfile for demo space: https://huggingface.co/spaces/MCP-1st-Birthday/agentic-economicsREADME.md→ submission documentation (this file)config.json→ Config file for the Covia grid venueactors- Convex Lisp actor code used for communication / contract makingprompts- Example prompts for doing different kings of transactions
The demo has dependencies on:
- Convex - Convex testnet code, used to run a separate Space for the shared global DLT layer at https://huggingface.co/spaces/mikera1337/convex-testnet
- Covia Venue - Grid venue code used for the grid server
Social Post (Public Announcement)
As part of our submission, we shared a social post explaining the concept and inspiration behind our Track 1 project.
Public Post: Post by Mike
This post introduces the idea of agents negotiating and executing economic contracts using MCP + Claude + Cursor + Covia Grid, and provides context for our submission.
Acknowledgements
Thanks to:
- HuggingFace and Partners for hosting the MCP hackathon
- The Anthropic team for providing the MCP server
- Anthropic, OpenAI, and Google for their APIs
- The MCP community for building a shared foundation
License
Apache 2.0 License.
See LICENSE.txt file in this repo.
