SciTeX MCP sunucusu

SciTeX (scitex)

A Python toolkit — AI for Science

SciTeX connects specialized scientific packages through a shared Python namespace, command-line tools, and interfaces for AI agents. Use individual packages for a focused task, or combine them in a research workflow.

• Deterministic research tools for literature, data, analysis, figures, and writing.
• Provenance-based verification with scitex-clew.
• Flexible interfaces: Python API, CLI, MCP, skills, and web API across the ecosystem.

Documentation · PyPI · SciTeX Cloud · Ecosystem

Problem and Solution

#	Problem	Solution
1	Fragmentation between literature, analysis, figures, and writing	Shared entry points backed by independent, specialized packages
2	Traceability gaps between reported results and their source data	Recorded provenance through scitex-clew and supported package integrations
3	Interfaces that differ across researchers and AI agents	Access through Python, CLI, MCP, skills, and Web APIs

Quick start

After installation, create a figure from a synthetic signal and save it with its FigRecipe recipe:

from pathlib import Path

import numpy as np
import scitex

x = np.linspace(0, 2 * np.pi, 200)
y = np.sin(x)

fig, ax = scitex.plt.subplots()
ax.plot(x, y)
ax.set_xyt("Time", "Amplitude", "Synthetic signal")

output = Path("outputs/signal.png").resolve()
scitex.io.save(fig, str(output))

Run the code in your project directory. It writes outputs/signal.png and the figure's recipe beside it. The absolute path makes the output location explicit.

Task	Python entry point	Owning package
Scientific file I/O	scitex.io	scitex-io
Statistics	scitex.stats	scitex-stats
Figures and recipes	scitex.plt	figrecipe
Literature	scitex.scholar	scitex-scholar
Manuscripts	scitex.writer	scitex-writer
Experiment sessions	scitex.session	scitex-session

Table 1. Selected Python entry points and their owning packages.

The umbrella delegates implementation to the owning packages. Their documentation describes detailed APIs, dependencies, and configuration.

Demo — SciTeX Cloud apps

SciTeX Hub provides the graphical workspace; specialized packages provide the apps. These screenshots show development builds with synthetic/demo content, not research findings.

Figure 1. FigRecipe: editable figures and recipes. Figure 2. Writer: manuscript editing and section previews. Captured November 2–3, 2026; image sources and hashes.

SciTeX Cloud is in alpha. Features and data formats may change. Keep backups.

Explore the apps · Watch the research demo and inspect its outputs

Installation

Create a virtual environment and install the toolkit:

uv venv .venv
source .venv/bin/activate
uv pip install scitex
scitex --help

Use the equivalent activation command for your shell. Python requirements are declared on PyPI. Optional integrations may have additional runtime or platform requirements.

Choose extras for the work you need, for example:

uv pip install "scitex[scholar,writer]"

See the installation guide for package and environment setup.

Architecture — Ecosystem

Role	Package
Python umbrella and integration	scitex — shared namespace and user entry points
Development tools and dispatchers	scitex-dev — discovery, shared checks, hooks, and skills
Standalone apps and Hub plugins	scitex-sdk — shared web/app primitives
Browser workspace and Web API hosting	scitex-hub
Domain logic and providers	Specialized leaf packages, each with its own APIs, documentation, and license

Table 2. Responsibilities across the ecosystem.

Leaves own their expertise; shared dispatchers discover and integrate their declared interfaces. Browse the ecosystem.

Five Interfaces

Interface	Entry point
Python API	import scitex
CLI	scitex --help, then the selected command's --help
MCP tools	scitex mcp --help; package tools are discovered from the installed environment
Skills	Packaged guidance and each leaf package's skills
Web API	Hub and app endpoints, served through scitex-hub and its plugins

Table 3. The five ecosystem interfaces; availability depends on the installed packages.

Available interfaces depend on installed packages, their versions, and configuration. Use documentation that matches your installed release.

For clients using an mcpServers configuration:

{
  "mcpServers": {
    "scitex": {
      "command": "/absolute/path/to/.venv/bin/scitex",
      "args": ["mcp", "start", "--transport", "stdio"]
    }
  }
}

Set command to the executable in your SciTeX environment. Inspect only the tool groups relevant to your task, for example:

scitex mcp list-tools --module stats

Provider credentials and external services are configured by the packages that use them.

In scitex 2.30.8, the skills CLI imports scitex_dev.skills, but its pinned scitex-dev 0.28.0 does not provide that module. Read the package-owned skill files directly for that release. The compatibility and interface updates are tracked in #387.

Provenance

Record Forward. Verify Backward.

scitex-clew records source data and inputs–executable–outputs relationships. Verification follows recorded dependencies back toward source data. Consult the owning package's integration and session guidance to enable the tracking appropriate to your workflow.

Documentation and contributing

• API reference
• Questions and ideas
• Bugs and feature requests
• Contributing guide and Contributor License Agreement

License

This package is licensed under AGPL-3.0, with commercial licensing described in the CLA. Check each leaf package's license for its terms.

Part of SciTeX

scitex is part of SciTeX, providing the Python umbrella and integration entry points.

Four Freedoms for Research

1. The freedom to run your research anywhere — your machine, your terms.
2. The freedom to study how every step works — from raw data to final manuscript.
3. The freedom to redistribute your workflows, not just your papers.
4. The freedom to modify any module and share improvements with the community.

AGPL-3.0 — because we believe research infrastructure deserves the same freedoms as the software it runs on.