scitex-python

SciTeX (scitex) exposes a rich MCP (Model Context Protocol) toolset to enable autonomous AI agents to orchestrate research workflows. The MCP tooling included with SciTeX allows agents to discover, invoke, and compose 120+ tools across categories like writer, scholar, capture, introspect, audio, stats, plt, diagram, dataset, and more. Typical usage involves starting the MCP server and querying or invoking tools through the MCP interface to perform tasks such as fetching papers, running statistical analyses, generating figures, and drafting manuscripts. The included example in the documentation demonstrates agent-driven workflows: a Scholar pass searches for literature, a Stats pass runs tests, a Plt pass creates figures, and a Writer pass compiles a manuscript, followed by a Capture pass for verification. This enables end-to-end research pipelines with reproducible traces and modular components.

To use the MCP capabilities, first install the SciTeX package, ensure you have Python and a compatible environment, then start the MCP server via scitex mcp start (as described in the MCP tooling reference). Once running, you can list tools, inspect APIs, and invoke specific MCP tools like scitex list-python-apis, scitex mcp list-tools, and scitex introspect api scitex.stats to plan and execute automated experiments. For agents, the MCP tools can be integrated into agent scripts or orchestration layers to perform autonomous, repeatable research tasks with minimal human intervention.

Prerequisites:

• Python 3.8+ and pip
• Network access to install Python packages

Installation steps:

1. Create and activate a virtual environment (recommended):

python3 -m venv env
source env/bin/activate  # on macOS/Linux
env\Scripts\activate     # on Windows

2. Install the SciTeX package (core):

pip install scitex

3. Install optional dependencies for plotting, statistics, and other modules (recommended):

pip install scitex[plt,stats,scholar]

4. Verify installation and MCP tooling availability:

scitex --version
scitex mcp list-tools

5. Start the MCP server (example):

scitex mcp start

6. (Optional) If you’re using uvx-based execution, ensure uvx is installed and available in your PATH:

pip install uvx
uvx --version

Notes and tips:

• The MCP toolset exposes many categories of tools (writer, scholar, capture, introspect, audio, stats, plt, diagram, dataset, social, canvas, template, verify, dev, ui, linter).
• You can customize tool behavior and environment via MCP server configuration. For CLI usage, you can inspect available tools with commands like scitex mcp list-tools and inspect APIs with scitex introspect api scitex.stats.
• Environment variables for credentials or configuration can be passed via an env block in the MCP config (e.g., to configure API keys or data sources).
• If you encounter tunneling or network issues in a corporate environment, ensure SSL/TLS is allowed and that the Python environment has access to required endpoints.
• When upgrading SciTeX, review the MCP tooling reference for any changes in tool IDs or categories, as tool names may be updated between releases.