# Hanzo MCP — 260+ tools and 1,800+ cloud operations for AI agents — Hanzo AI

> Model Context Protocol server for AI agents: filesystem, shell, AST code intelligence, Kai sub-10ms decisions, browser CDP, issue tracker, persistent memory, and 126 Hanzo Cloud services over stdio or streamable HTTP.

Hanzo MCP · 260+ tools · 126 cloud services

# The complete MCP runtime for coding, decisions, browser, and cloud

Give any model a filesystem, shell, semantic AST code index, Kai sub-10ms decisions, Playwright & Chrome DevTools browser control, issue tracking, persistent memory, and 1,800+ typed Hanzo Cloud operations over stdio or streamable HTTP.

[Browse 260+ tools](https://docs.hanzo.ai/docs/mcp-tools/all-tools)[Read the docs](https://docs.hanzo.ai/docs/mcp)[View on GitHub](https://github.com/hanzoai/mcp)

Open source, MIT. Identical tool names and schemas in TypeScript, Python, and Rust — plus remote streamable HTTP at api.hanzo.ai/v1/mcp.

Available throughClaude DesktopClaude CodeCursorHanzo DevAny MCP client

38 unified action tools + 1,800+ Cloud operations + 260+ specialized servers

01 / 03

Full-stack agent capabilities, ready out of the boxCore runtime (fs, exec, code, git, fetch, workspace, ui), semantic code intelligence (code_search, code_context, code_ask, code_index), Kai decision & safety guardrails (kai_decide, kai_command_guard, kai_injection_screen), browser CDP, issue tracker, and persistent memory.

126 Hanzo Cloud services, 1,800+ typed operationsThrough the unified hanzo router and /docs/mcp-tools catalog, agents provision sandboxes, deploy containers across 34 regions, query vector indexes, rotate KMS secrets, verify IAM policies, and inspect billing usage under one bearer key.

Scoped permissions & Kai safety screeningEvery shell command and external input can pass through kai_command_guard and kai_injection_screen in single-digit milliseconds. Enable or drop tool groups per environment with --enable-tools and --disable-tools.

- Full-stack agent capabilities, ready out of the box
- 126 Hanzo Cloud services, 1,800+ typed operations
- Scoped permissions & Kai safety screening

Paste and ship

Paste this into any agent. It reads the skill manifest and calls MCP from there.

Agent promptCopy

```
Read https://hanzo.ai/skill.md and use Hanzo MCP in my project. Start with: hanzo mcp create
```

Tool domains

## Every surface an autonomous team needs

From local AST refactoring and headless browser verification to multi-region cloud deployments.

### fs & exec

read, write, stat, list, mkdir, rm, mv, apply_patch, search_text — plus exec with run, background, ps, kill, and logs so long-running dev servers and test suites stream cleanly.

### code_search, code_context, code_ask, code_index

AST-aware symbol resolution, call graphs, cross-repository indexing, and structural transforms. Find every caller of a symbol in 2ms without grep noise.

### kai_decide, kai_command_guard, kai_injection_screen

13 purpose-built Kai decision, scoring, routing, log triage, recovery, and context compaction tools running in 6.4ms at a fraction of LLM token cost.

### browser, cdp, git & fetch

Full Playwright and Chrome DevTools Protocol automation (navigate, click, screenshot, console, network), complete git workflows, and streaming HTTP fetch.

### tracker, memory, think, plan & tasks

tracker_boards, tracker_issues, tracker_create, and tracker_update sync directly with engineering boards, while memory, think, and plan persist architecture context across sessions.

### hanzo (126 Cloud services · 1,800+ ops)

Direct typed access to Deploy, Sandbox, IAM, KMS, KV, S3, Vector, Search, Crawl, Workflow, Commerce, and Billing — inspectable at docs.hanzo.ai/docs/mcp-tools.

Get started

## One config block

Use the --package= form for local stdio, or point any MCP client directly at https://api.hanzo.ai/v1/mcp over streamable HTTP.

json

```
// .mcp.json — add Hanzo MCP to any MCP client { "mcpServers": { "hanzo": { "command": "npx", "args": ["-y", "hanzo", "mcp"] } } } // Or install the whole native stack once and run it yourself: // npm i -g hanzo // hanzo mcp # stdio; the binary downloads on first run // hanzo mcp -t sse # or over HTTP
```

[All MCP Tools Reference](https://docs.hanzo.ai/docs/mcp-tools/all-tools)[View on GitHub](https://github.com/hanzoai/mcp)

## Add Hanzo MCP to your client

One block in .mcp.json, or connect over streamable HTTP at api.hanzo.ai/v1/mcp.

[Browse All Tools](https://docs.hanzo.ai/docs/mcp-tools/all-tools)[GitHub](https://github.com/hanzoai/mcp)[Hanzo Platform](https://platform.hanzo.ai)

MCP

[MCP docs →](https://docs.hanzo.ai/docs/mcp)

## Get MCP

Model Context Protocol — 260+ tools

License: Apache-2.0[hanzoai/mcp](https://github.com/hanzoai/mcp)17 stars

[Deploy to Cloud](https://platform.hanzo.ai/deploy)[Self-host](https://docs.hanzo.ai/docs/mcp)
