This subnet brings MCP (Model Context Protocol) servers into the Bittensor ecosystem, enabling AI models to securely interact with external tools, data sources, and execution environments.
By combining MCP with Bittensor's incentive-driven design, the subnet creates a competitive environment where miners are rewarded for delivering high-availability, low-latency, and high-quality MCP services.
Our goal is to build a decentralized platform of production-ready MCP servers that:
- extend AI models with real-world capabilities through standardized tool interfaces
- enable seamless integrations across systems, databases, and APIs
- support businesses, individual users, and other Bittensor subnets with reliable infrastructure
- continuously improve through competition, real usage metrics, and community feedback
The subnet aims to become the standard for decentralized AI tooling infrastructure - a universal platform designed for the entire Bittensor ecosystem and beyond. We're building flexible, production-ready MCP servers that any subnet, developer, or business can leverage, where quality and performance are rewarded through market-driven incentives.
The Platform serves as the central orchestration layer of the subnet, responsible for:
- Service Management: Hosting and ensuring availability of MCP services
- Algorithm Registry: Managing and collecting models/algorithms submitted by miners
- Analytics Dashboard: Providing real-time performance metrics and insights
- Miner Registration: Processing and validating new miner submissions
- Validation Orchestration: Coordinating task distribution across validators
- Quality Assessment: Final evaluation of miner performance based on validator data
Validators score miner solutions by:
- Fetching execution results from the platform
- Evaluatimg solution quality based on competition criteria
- Reporting scores to the platform for weight calculation
Min Hardware Requirements:
- 4 CPU cores
- 16 GB RAM
- 200 GB SSD storage
On SOMA, any problem that can be meaningfully solved using an MCP server - and that can significantly improve agent performance - may become a competition target. Miners compete to deliver the most effective model or algorithm for a given task.
The miner’s responsibility is to design and implement model or algorithm that solves the defined problem as effectively as possible and upload it to the platform
All a miner needs to participate is:
- A working algorithm that solves the active MCP task
- A registered hotkey on netuid 114
The platform handles orchestration and evaluation. Validators automatically retrieve submitted solutions associated with registered hotkeys and score them according to the active competition criteria.
The inaugural MCP challenge focuses on context compression - a critical problem for reducing costs and improving quality in agentic AI systems.
- Lower inference costs
- Faster responses
- Longer effective memory windows
- Stronger reasoning from distilled context
- Scalable intelligence for multi-agent systems
The subnet operates on a recurring weekly competition cycle using a winner-takes-all incentive model.
The platform aggregates validator assessments into final rankings, and validators set on-chain weights based on platform-calculated scores.
Each miner may upload one algorithm per hotkey per competition cycle.
The primary objective of the incentive mechanism is to reward:
- State-of-the-art algorithmic quality
- Solutions aligned with best engineering practices
- Robust, generalizable approaches rather than overfitted heuristics
- Meaningful innovation that advances the field
Each competition lasts one week and consists of three distinct phases:
- Miners upload or update their algorithms on the platform
- Submissions must be associated with a registered hotkey
- Only code submitted during this window is eligible for the current cycle
- Submitted solutions undergo automated validation and integrity checks
- The platform evaluates baseline performance and technical correctness
- A subset of top-performing qualified submissions is selected to advance to the main competition phase
This screening stage ensures:
- Stability and security of evaluated solutions
- Minimum quality standards
- Efficient allocation of validator resources
- Qualified solutions are evaluated continuously under live competition conditions
- Validators score miners according to the active task criteria
- Final rankings are computed at the end of the weekly cycle
- The top-ranked miner is declared the weekly winner
- The winner receives the subnet incentive allocation
- After the reward period ends, a new weekly competition cycle begins
This structure:
- Encourages rapid iteration and continuous improvement
- Prevents stagnant dominance without sustained performance
- Ensures only stable and high-quality solutions reach the competitive stage
- Creates predictable and recurring incentive cycles
SOMA is designed for seamless integration with other Bittensor subnets.
Coming Soon: Details about confirmed partnerships and integrations across subnets.
This subnet is community-driven and values transparency and collaboration:
- Cross-Subnet Partnerships: Actively seeking integrations with complementary subnets
- Community Voting: Future governance model for selecting new MCP server types
- Bug Bounty Program: Planned initiative to ensure validation integrity and security
- Open Development: Public roadmap and regular updates on subnet evolution
- Join our Discord for technical discussions
- Contribute to the GitHub repository
- Propose new MCP server types and use cases
- Participate in testing and feedback cycles
- Discord: Join our server
- Twitter: @SomaSubnet
- GitHub: github.com/DendriteHQ/SOMA
- Email: thesoma@dendrite.holdings
Built on Bittensor | Powered by the Community