Colber MCP Server
Exposes Colber's five integrated services (Reputation, Memory, Observability, Negotiation, Insurance) as 27 MCP tools for AI agents, enabling trust, coordination, and continuity.
README
Colber
Trust, coordination & continuity — for the agent economy. Five integrated services AI agents need to operate at scale: Reputation · Memory · Observability · Negotiation · Insurance — exposed via SDKs (TypeScript, Python) and the Model Context Protocol.
🌐 https://colber.dev · 📦 npm @colber/* · 🐍 PyPI colber-sdk · 🔌 MCP-native
What is Colber?
Colber is the infrastructure layer of trust, coordination and continuity for the agent economy. The hosted platform at https://colber.dev exposes five integrated capabilities through one consistent surface (REST · gRPC · MCP):
| Module | What it does |
|---|---|
| Reputation | Cryptographic reputation oracle. DID-based scoring with offline-verifiable attestations (Ed25519 + JCS RFC 8785). |
| Memory | Persistent semantic memory with vector search, ACLs, and opt-in encryption. |
| Observability | Distributed A2A tracing and logging. ClickHouse-backed, OpenTelemetry-compatible. |
| Negotiation | Multi-party broker with auctions, multi-criteria, and signed settlement (event-sourced). |
| Insurance | Deliverable guarantees: pricing by reputation, escrow, claim arbitration. |
Plus an identity support service (DID:key + Ed25519 signature verification) used by every module.
🏗️ Status — v1 shipped. All five modules + identity are live on
https://colber.dev, end-to-end tested (23/23 green), with 27 MCP tools published.
What's in this repo (open core)
This repository is the public open-core of Colber. It contains the code you need to integrate with Colber from your own agent, plus the public protocol contract:
apps/
├── sdk-typescript/ → npm @colber/sdk
├── sdk-python/ → PyPI colber-sdk
├── mcp-server/ → npm @colber/mcp (CLI: npx -y @colber/mcp)
└── site/ → https://colber.dev (landing source)
packages/
├── core-types/ → Public protocol types (errors, envelopes, DIDs)
├── core-crypto/ → Client-side crypto helpers (DID:key, Ed25519, JCS canonicalisation)
├── core-config/ → Env-var validation utilities (zod schemas)
└── core-logger/ → Structured logging utilities (pino + traceId)
tooling/ → Shared TS / ESLint configs
.github/ → Issue + PR templates
docs/diagrams/ → High-level architecture diagrams (Mermaid)
docs/MCP_REGISTRIES.md → Submission templates for Anthropic, Smithery, mcp.so
Everything in this repo is Apache-2.0. You can fork it, embed it in your products, ship modified versions of the SDKs, contribute back via PR.
Not in this repo
The server-side implementation of the five modules + identity (the actual Reputation engine, Memory vector index, Observability ingestion, Negotiation event store, Insurance escrow logic, operator console) is proprietary and runs on https://colber.dev. To use it, you call the hosted endpoints from the SDKs or the MCP server. This is the standard open-core model used by Stripe, Datadog, Auth0.
Quick start
TypeScript
npm install @colber/sdk
import { ColberClient } from '@colber/sdk';
const colber = new ColberClient({ baseUrl: 'https://api.colber.dev' });
const score = await colber.reputation.score('did:key:z6Mk...');
console.log(score);
Python
pip install colber-sdk
from colber import ColberClient
colber = ColberClient(base_url="https://api.colber.dev")
score = colber.reputation.score("did:key:z6Mk...")
print(score)
MCP — Claude Desktop / Claude Code / Cline / Continue
Add to your MCP client configuration (e.g. ~/Library/Application Support/Claude/claude_desktop_config.json on macOS):
{
"mcpServers": {
"colber": {
"command": "npx",
"args": ["-y", "@colber/mcp"]
}
}
}
You instantly get 27 Colber tools — reputation lookups, memory search, signed feedback, multi-party negotiations, insurance quotes, and more — directly available to your AI assistant.
See apps/mcp-server/README.md for full configuration options (HTTP transport, custom backend URLs, auth tokens).
Verifying reputation attestations offline
One of Colber's strongest properties: every reputation score comes with a cryptographic attestation that can be verified without contacting Colber, using only the platform's public key.
import { ColberClient } from '@colber/sdk';
import { verifyAttestation, COLBER_PLATFORM_PUBLIC_KEY } from '@colber/sdk/crypto';
const colber = new ColberClient({ baseUrl: 'https://api.colber.dev' });
const score = await colber.reputation.score('did:key:z6Mk...');
const valid = await verifyAttestation(score, COLBER_PLATFORM_PUBLIC_KEY);
// `valid` is true iff the score was actually emitted by Colber.
The verification logic lives in packages/core-crypto/ — fully open, auditable, reproducible. You don't have to trust our server to trust the score.
Standards we speak
Colber is built on top of open standards rather than reinventing them:
- MCP — Model Context Protocol (
@colber/mcpships 27 tools) - A2A — Agent-to-Agent observability
- DID — W3C Decentralized Identifiers (
did:key, Ed25519 multibasez6Mk…) - JCS RFC 8785 — JSON canonicalization for signed payloads
- Ed25519 — signatures (via
@noble/ed25519) - OpenTelemetry — observability export (planned)
- EIP-712 — on-chain signatures (planned for INSURANCE on-chain variant)
Architecture overview
flowchart TB
classDef sdk fill:#F8FAFC,stroke:#475569,color:#0F172A
classDef hosted fill:#EFF4FF,stroke:#1E3A8A,color:#1E3A8A
classDef this fill:#FFFBEB,stroke:#A16207,color:#78350F
User["🧑 Your agent / app"]:::sdk
SDK["@colber/sdk · colber-sdk"]:::this
MCP["@colber/mcp"]:::this
User --> SDK
User --> MCP
SDK -->|HTTPS| API
MCP -->|HTTPS| API
subgraph hosted["Colber hosted platform — colber.dev"]
API["api.colber.dev (REST · gRPC)"]:::hosted
REP["Reputation"]:::hosted
MEM["Memory"]:::hosted
OBS["Observability"]:::hosted
NEG["Negotiation"]:::hosted
INS["Insurance"]:::hosted
ID["Identity"]:::hosted
API --> ID
API --> REP
API --> MEM
API --> OBS
API --> NEG
API --> INS
end
For the high-level functional architecture, see docs/diagrams/.
Local development
This repo is a Turborepo + pnpm workspace.
Prerequisites
- Node.js 22+ (
.nvmrc) - pnpm 9.12+ (
corepack enable && corepack prepare pnpm@9.12.3 --activate)
Install + checks
git clone https://github.com/Obi49/Colber.git
cd Colber
pnpm install
pnpm typecheck # 11/11 green
pnpm test # 11/11 green
pnpm lint # 0 errors, 0 warnings
pnpm build # 7/7 green
Working on the SDK
pnpm --filter @colber/sdk dev # watch build
pnpm --filter @colber/sdk test:watch # watch tests
Running the MCP server locally
pnpm --filter @colber/mcp build
node apps/mcp-server/dist/server.js # stdio (default)
node apps/mcp-server/dist/server.js --transport=http --port=14080
Running the landing locally
pnpm --filter @colber/site dev
# → http://localhost:3001
Contributing
We welcome contributions to the open-core surface — SDKs, MCP server, public types, the website, and documentation. See CONTRIBUTING.md for the workflow (Conventional Commits, DCO, no --no-verify).
For security issues, please follow SECURITY.md — do not file public issues.
License
Apache License 2.0 — see NOTICE for attribution and project history (the project was previously named AgentStack, then Praxis, before being renamed Colber in May 2026).
The hosted services on colber.dev are operated under separate commercial terms; using them is subject to the Colber Terms of Service (link forthcoming).
Author
Johan / Colber — dof1502.mwm27@gmail.com
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。
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。
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。
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。