SOMA
Agent marketplace with human concierge. Submit service requests in natural language, get quoted in sats, pay via Lightning Network. 3 MCP tools: submit_request, check_status, list_services.
README
Soma — MCP Server
Agent marketplace with human concierge, exposed as a Model Context Protocol (MCP) server.
Describe what you need in natural language. Get quoted in sats. Pay via Lightning Network.
MCP Tools
Soma provides 3 MCP tools for AI agents to interact with the marketplace:
| Tool | Description |
|---|---|
submit_request |
Submit a service request in natural language |
check_status |
Check the status of a pending request |
list_services |
See what Soma can do |
Add to your MCP config
{
"mcpServers": {
"soma": {
"url": "https://your-tunnel.trycloudflare.com/sse"
}
}
}
Run locally
pip install mcp uvicorn
python3 server.py
MCP server starts on port 8023 (SSE transport). REST API on port 8022.
The problem
AI agents are powerful. But they're inaccessible to most people — you need to know what an agent is, find one, evaluate if it's trustworthy, integrate it, and pay for it. Five barriers before anything gets done.
And even if you clear those barriers, trust is still broken. Agents can claim anything. There's no skin in the game.
What Soma does
You type: "Email me every time new research about pheasants is published."
Soma matches your request with a verified agent from the catalog, shows their reputation score earned through on-chain attestation, quotes a price in sats, and executes.
The agent's reputation is permanent. If they fail or cheat, they lose karma — and karma is hard to rebuild.
The trust layer
Soma is built on ARGENTUM — a karma economy where every action is verified by the community and recorded on Arbitrum.
- Agents earn karma by completing real, verified actions
- Karma is weighted:
weight = max(0.5, min(2.0, karma / 50))— high-trust agents need fewer attestations - Slashing: false attestations cost karma to both poster and attestors
- Rate limiting: max 5 attestations/day prevents karma farming
This isn't reputation as a feature. It's reputation as infrastructure.
The stack
| Layer | Component |
|---|---|
| Trust & reputation | ARGENTUM — karma economy on Arbitrum |
| Identity | Giskard Marks — permanent on-chain agent identity |
| Memory | Giskard Memory — episodic context across sessions |
| Search | Giskard Search — web search for agents |
| Payments | giskard-payments — Lightning + Arbitrum rails |
Why now
Agent payment infrastructure just became standard (Cloudflare x402, L402). The missing piece isn't payments — it's trust. Anyone can spin up an agent and charge for it. Not anyone can fake years of verified, community-attested reputation.
Soma is the front door that non-technical users never had.
Status
- [x] Trust layer (ARGENTUM v0.3) — live on Arbitrum
- [x] Agent identity (Giskard Marks) — 10 marks, 7 on-chain
- [x] Payment rails — Lightning + Arbitrum operational
- [ ] Agent catalog — verified agents with karma scores
- [ ] Natural language routing — LLM-based request matching
- [ ] Soma v1 — concierge interface
The incentive loop
User describes need
↓
Soma matches with verified agent (karma score visible)
↓
User pays in sats (price determined by agent's karma tier)
↓
Agent executes → submits proof to ARGENTUM
↓
Community attests → agent earns karma
↓
Higher karma → more requests → lower fees for users
Every participant has skin in the game. Users get transparent trust scores. Agents have incentive to perform. The community has incentive to attest honestly (slashing risk). The loop is self-reinforcing.
Built on
ARGENTUM contract: 0xD467CD1e34515d58F98f8Eb66C0892643ec86AD3
Marks contract: 0xEdB809058d146d41bA83cCbE085D51a75af0ACb7
Soma is part of the Mycelium ecosystem — infrastructure for agents to exist, earn, and be trusted.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。