Files
zooqueenandhanzo-dev aac169347a feat(tools): research action on net/fetch — cited answers over /v1/ask
Search finds pages, web_read reads one, research is the whole loop behind a
single door: the live cloud answer engine (api.hanzo.ai /v1/ask, mode=research)
plans queries, searches, reads the best pages and writes an answer that cites
them. It is an action on the same fetch tool, not a new package — search, read
and research are one concern (the web) and share one client and one auth.

The engine replies as server-sent events, which is the cloud's other reply
shape, so HanzoCloud gains stream() beside get/post: the base URL, the auth
header and the error mapping stay in the one place that already owns them.
Each `data:` frame is yielded as a typed event (status | sources | text |
follow_ups | done | error); the terminal [DONE] sentinel ends the stream and is
never an event. Frames accumulate into {answer, sources, follow_ups}; `deep`
normalizes to `research` (one pass, two names) and any other mode is refused
rather than silently downgraded.

hanzo-tools 0.3.4->0.3.5, -net 0.1.3->0.1.4, and -net's floor moves to
hanzo-tools>=0.3.5 so the action cannot install against a client with no
stream(). Tests replay a canned SSE stream through the real parser.

Co-authored-by: Hanzo Dev <dev@hanzo.ai>
2026-08-04 22:48:19 -07:00
..

hanzo-tools

Core infrastructure and plugin framework for Hanzo AI's modular MCP tool system.

Install

pip install hanzo-tools          # Core only
pip install hanzo-tools[dev]     # filesystem, shell, editor, lsp, refactor, todo, reasoning, config
pip install hanzo-tools[ai]      # llm, agent, memory
pip install hanzo-tools[all]     # Everything

Individual tool packages can be installed separately:

pip install hanzo-tools-fs
pip install hanzo-tools-shell
pip install hanzo-tools-browser
pip install hanzo-tools-llm
pip install hanzo-tools-memory
pip install hanzo-tools-editor
pip install hanzo-tools-vector
# ... etc

Usage

Register all discovered tools

from hanzo_tools import discover_tools, register_all
from mcp.server import FastMCP

mcp = FastMCP("my-server")

# Auto-discover and register all installed tool packages
registered = register_all(mcp)

Register individual tool packages

from hanzo_tools.fs import register_tools as register_fs
from hanzo_tools.shell import register_tools as register_shell

register_fs(mcp, permission_manager)
register_shell(mcp)

Build a custom tool

from typing import Any
from mcp.server import FastMCP
from mcp.server.fastmcp import Context as MCPContext
from hanzo_tools.core import BaseTool, ToolRegistry, with_error_logging

class WeatherTool(BaseTool):
    @property
    def name(self) -> str:
        return "get_weather"

    @property
    def description(self) -> str:
        return "Get current weather for a location"

    @with_error_logging("get_weather")
    async def call(self, ctx: MCPContext, **params: Any) -> str:
        location = params["location"]
        return f"Weather in {location}: 72°F, sunny"

    def register(self, mcp_server: FastMCP) -> None:
        @mcp_server.tool()
        async def get_weather(location: str, ctx: MCPContext) -> str:
            """Get current weather for a location."""
            return await self.call(ctx, location=location)

# Register with the MCP server
ToolRegistry.register_tool(mcp, WeatherTool())

Filesystem tools with permissions

from hanzo_tools.core import FileSystemTool, PermissionManager

pm = PermissionManager(
    allowed_paths=["/home/user/projects"],
    deny_patterns=[".git", "node_modules", ".env"],
)

class ReadFileTool(FileSystemTool):
    def __init__(self):
        super().__init__(permission_manager=pm)

    @property
    def name(self) -> str:
        return "read_file"

    @property
    def description(self) -> str:
        return "Read a file"

    async def call(self, ctx, **params):
        path = params["path"]
        if not self.is_path_allowed(path):
            return "Access denied"
        return open(path).read()

    def register(self, mcp_server):
        @mcp_server.tool()
        async def read_file(path: str, ctx) -> str:
            """Read a file."""
            return await self.call(ctx, path=path)

Decorators

from hanzo_tools.core import auto_timeout, with_error_logging, handle_connection_errors

@auto_timeout("search", timeout=120)
@with_error_logging("search")
@handle_connection_errors
async def search_files(pattern: str, path: str) -> str:
    ...

Timeouts are configurable via environment variables:

HANZO_TIMEOUT_SEARCH=300   # Override search timeout to 5 minutes
HANZO_TIMEOUT_BROWSER=600  # Override browser timeout to 10 minutes

Tool enable/disable

from hanzo_tools.core import ToolRegistry

# Disable a tool at runtime
ToolRegistry.set_tool_enabled("browser", False)

# Check if a tool is enabled
if ToolRegistry.is_tool_enabled("browser"):
    ...

Tool states persist to ~/.hanzo/mcp/tool_states.json.

Architecture

hanzo-tools (core)
├── BaseTool          — Abstract base for all tools
├── FileSystemTool    — Base for filesystem tools with permissions
├── ToolRegistry      — Enable/disable and registration
├── ToolContext       — Execution context with logging/progress
├── PermissionManager — Path-based access control
├── MCPResourceDocument — Structured response formatting
├── auto_timeout      — Configurable async timeouts
├── with_error_logging — Error logging to ~/.hanzo/mcp/logs/
├── handle_connection_errors — Graceful disconnect handling
├── validate_path_parameter  — Path validation
├── discover_tools()  — Plugin discovery via entry points
└── register_all()    — Auto-register all discovered tools

hanzo-tools-fs        — File read/write/search/glob
hanzo-tools-shell     — Shell command execution
hanzo-tools-browser   — Browser automation (Playwright)
hanzo-tools-editor    — Code editing with AST awareness
hanzo-tools-lsp       — Language Server Protocol integration
hanzo-tools-refactor  — Refactoring operations
hanzo-tools-llm       — LLM inference tools
hanzo-tools-agent     — Agent orchestration
hanzo-tools-memory    — Persistent memory/context
hanzo-tools-vector    — Vector DB operations
hanzo-tools-database  — Database query tools
hanzo-tools-jupyter   — Jupyter notebook tools
hanzo-tools-todo      — Task/todo management
hanzo-tools-reasoning — Chain-of-thought reasoning
hanzo-tools-config    — Configuration management
hanzo-tools-mcp       — MCP protocol utilities
hanzo-tools-computer  — Computer use (screen/keyboard)

Tool packages register via importlib.metadata entry points (group="hanzo.tools"), enabling automatic discovery without explicit imports.

API

Core Classes

Class Purpose
BaseTool Abstract base — implement name, description, call(), register()
FileSystemTool Extends BaseTool with permission_manager and validate_path()
ToolRegistry Class-level enable/disable with JSON persistence
ToolContext Wraps MCP context with info(), warning(), error(), progress()
PermissionManager Allowed paths + deny patterns for filesystem access control
MCPResourceDocument Response type with to_json_string(), to_readable_string(), to_dict()
ValidationResult Boolean result with optional error message

Decorators

Decorator Purpose
@auto_timeout(name, timeout=None) Async timeout with env var override (HANZO_TIMEOUT_*)
@with_error_logging(name) Log errors to ~/.hanzo/mcp/logs/ and return friendly messages
@handle_connection_errors Catch disconnects gracefully
@retry(max_attempts, delay, backoff) Exponential backoff retry

Top-level Functions

Function Purpose
discover_tools() Find all installed tool packages via entry points
register_all(mcp, pm, enabled) Register all discovered tools with an MCP server

License

MIT