mirror of
https://github.com/zoo-apps/ZIPs.git
synced 2026-10-30 14:57:22 +00:00
zeekayandHanzo Dev
d1410729da
Bylines, paper authorship, correspondence addresses and the founder blurb under each post named an individual. They name the organisation that published the work instead: the brand on posts and pages, the research byline on papers, and research@ in place of a personal address. Both are what these repositories already used elsewhere — nothing new was coined. Co-authors are untouched. Where a paper named someone else alongside, they stay named; only the one byline changed. Bibliography entries, running heads, PDF metadata and affiliations move with it, so a citation still resolves to whoever the cited paper names. Sample and fixture data uses the placeholder name these tests already use elsewhere, so no assertion changed meaning. Co-authored-by: Hanzo Dev <[email protected]>
5.1 KiB
5.1 KiB
zip, title, description, author, authors, status, type, category, originated, traces-from, follow-on, created, tags, requires, references, repository, license
| zip | title | description | author | authors | status | type | category | originated | traces-from | follow-on | created | tags | requires | references | repository | license | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0412 | MCP Server Architecture | Model Context Protocol server architecture enabling AI models to use 260+ tools for code, data, web, and infrastructure interaction | Zoo Labs Foundation |
|
Final | Standards Track | AI | 2023-12 | ZIP-0400 / Whitepaper Section 08 |
|
2023-12-01 |
|
|
MCP-Spec | https://github.com/hanzoai/mcp | CC BY 4.0 |
ZIP-0412: MCP Server Architecture
Abstract
This proposal defines the architecture for Hanzo's Model Context Protocol (MCP) server infrastructure, providing 260+ tools that AI models can invoke to interact with code, data, web services, and infrastructure. MCP transforms language models from passive text generators into active agents capable of reading files, executing code, querying databases, browsing the web, managing containers, and orchestrating complex multi-step workflows. This is the tool-use infrastructure that powers all Hanzo agent systems.
Motivation
The conversational AI system (ZIP-0400) and agent framework (ZIP-0402) require models to take actions in the world, not just generate text. When a conservation agent identifies that a species range has shifted, it needs to:
- Query the latest satellite imagery (tool: fetch)
- Run a habitat analysis model (tool: exec)
- Update the species database (tool: write)
- Notify relevant conservation organizations (tool: email)
- Create a conservation alert (tool: api)
Without standardized tool access, each of these would require custom integration code. MCP provides a universal protocol for model-tool interaction.
Specification
Architecture
AI Model (Zen family)
│
│ MCP Protocol (JSON-RPC over stdio/SSE/WebSocket)
v
┌──────────────────────┐
│ MCP Server │
│ │
│ Tool Registry │
│ ├── Code Tools │ read, write, exec, ast, git, lsp
│ ├── Data Tools │ fetch, search, jq, sql, vector
│ ├── Web Tools │ browser, curl, wget, scrape
│ ├── Infra Tools │ docker, k8s, deploy, monitor
│ ├── AI Tools │ llm, embed, generate, transcribe
│ └── Domain Tools │ conservation, species, habitat
│ │
│ Permission Layer │
│ Context Manager │
│ Rate Limiter │
└──────────────────────┘
Tool Categories
| Category | Tools | Count | Description |
|---|---|---|---|
| Code | read, write, edit, exec, ast, git, lsp, refactor, review | 40+ | Source code manipulation |
| Data | fetch, search, jq, sql, vector, csv, json | 30+ | Data retrieval and transformation |
| Web | browser, curl, wget, scrape, screenshot | 20+ | Web interaction |
| Infrastructure | docker, k8s, deploy, monitor, logs | 25+ | DevOps and infrastructure |
| AI | llm, embed, generate, transcribe, translate | 15+ | AI model invocation |
| Communication | email, slack, webhook, notify | 10+ | External communication |
| Conservation | species, habitat, iucn, satellite, camera-trap | 20+ | Domain-specific |
| Utility | think, memory, todo, mode, version | 100+ | Agent utilities |
Protocol
MCP uses JSON-RPC 2.0 over three transport options:
- stdio: For local tool execution (lowest latency)
- SSE: For server-sent events (streaming results)
- WebSocket: For bidirectional real-time communication
Permission Model
Tools are gated by a capability-based permission system:
- Read-only tools: Available to all agents by default
- Write tools: Require explicit user consent per session
- Destructive tools: Require per-invocation approval
- Infrastructure tools: Require org-level authorization
Context Management
MCP manages the model's context window by:
- Compressing tool outputs that exceed token budgets
- Caching frequently-used tool results
- Prefetching likely-needed context based on conversation trajectory
Research Papers
- hanzo-agent-sdk -- Agent SDK with MCP integration
- zen-agent-framework -- Zen agent framework architecture
- zen-agent -- Zen-Agent model with native tool use
Implementation
- hanzo/mcp: Production MCP server with 260+ tools (
npm install -g @hanzo/mcp) - hanzo/agent: Multi-agent SDK with MCP client integration
- hanzo/operative: Computer use framework built on MCP (ZIP-0422)
- hanzo/chat: Chat interface with MCP tool invocation
Timeline
- Originated: December 2023 (MCP protocol design)
- Research:
hanzo-agent-sdkpublished 2024 - Implementation: Hanzo MCP server with 260+ tools deployed 2024