CLI¶

The toolschema command-line tool inspects, compares, and exports schemas.

pip install toolschema
toolschema --help

`inspect`¶

Print tool JSON for a single function.

toolschema inspect myapp.tools:search_github
toolschema inspect myapp.tools:search_github --format mcp
toolschema inspect myapp.tools:search_github --format openai,mcp,anthropic

Target syntax¶

Format	Example
Module function	`mypackage.tools:add`
Nested attribute	`mypackage.registry:TOOLS.search`

Formats¶

`--format` value	Output
`canonical`	`ToolDefinition.to_json_schema()` (default)
`openai`	OpenAI function tool
`openai-strict`	OpenAI with `strict=True`
`anthropic`	Anthropic tool
`mcp`	MCP tools/list entry
`gemini`	Gemini FunctionDeclaration

Multiple formats: comma-separated, no spaces (or spaces trimmed).

`diff`¶

Compare adapter outputs for the same function.

toolschema diff myapp.tools:add --targets openai,mcp
toolschema diff myapp.tools:add --targets openai,openai-strict,mcp,anthropic

Prints JSON with equal flags and key paths for each pair.

`export`¶

Export schemas for all callables in a module.

toolschema export myapp.tools
toolschema export myapp.tools --output tools.json

Skips objects where schema() raises TypeError or ValueError.

Output shape:

{
  "module": "myapp.tools",
  "tools": [
    { "name": "...", "description": "...", "parameters": {...}, "_source": "myapp.tools:add" }
  ]
}

`init`¶

Scaffold a new MCP server project.

toolschema init my-mcp-server
toolschema init my-mcp-server --path /tmp

Creates:

my-mcp-server/
├── pyproject.toml
├── README.md
├── claude_desktop_config.example.json
└── src/my_mcp_server/
    ├── tools.py
    └── __main__.py

Next steps printed by CLI:

cd my-mcp-server
uv sync
uv run python -m my_mcp_server --check
uv run python -m my_mcp_server

Exit codes¶

Code	Meaning
`0`	Success
`1`	Runtime error (import failure, bad target, etc.)
`2`	Argument error

Run as module¶

python -m toolschema inspect myapp.tools:add --format mcp

CLI¶

inspect¶