Mathematica MCP

A front-end / notebook automation layer for Mathematica, built for AI agents.

Your AI agent can write Mathematica code. This server lets it run that code in your live Mathematica session: create and edit notebooks, build interactive Manipulate panels, capture screenshots, verify derivations step by step, and read .nb files even without a kernel. Works with Claude, Cursor, VS Code, Codex, and Gemini.

It runs beside the official Wolfram Local MCP, not instead of it: Wolfram's server is the reference evaluator and documentation surface; this one owns the live notebook. See How it compares.

Upgrading from an earlier version? See the Migration Guide.

Watch it in action

An AI agent solving math, generating plots, and controlling a live Mathematica notebook. Errors are returned directly to the agent, no copy-pasting notebook output back into chat.

Why this exists

LLMs can write Wolfram Language, but they cannot run it, see the result, or fix their own mistakes. This server closes that loop:

• Live notebook control: the agent creates, edits, evaluates, and screenshots real notebooks in your running Mathematica front end. Interactive content (Manipulate, Dynamic, sliders) renders as live panels.

• Fast by default: computation runs on a persistent kernel session that starts warming the moment the server launches; calls return in milliseconds, not the ~13 seconds of a cold wolframscript start per request.

• Self-debugging agents: Mathematica errors flow back with a suggested_fix and, when derivable, a ready-to-run retry_with call. No copying red text from the notebook into chat.

• Derivation checking: verify_derivation validates a chain of mathematical steps and pinpoints the first invalid one.

• Reads notebooks without a license: read_notebook_file parses .nb files in pure Python when no kernel is available (a kernel is used for higher fidelity when present).

• Local and private: core execution stays on your machine. Cloud services are contacted only by opt-in tools like wolfram_alpha.

Related MCP server: JupyterMCP

What you can ask for

"Integrate x^2 sin(x) from 0 to pi, then verify the result."

evaluate("Integrate[x^2 Sin[x], {x, 0, Pi}]")   =>  -4 + Pi^2
verify_derivation(steps=["Integrate[x^2 Sin[x], {x, 0, Pi}]", "-4 + Pi^2"])
=> Step 1 → 2: ✓ VALID
   All steps are valid!

"Plot the sombrero function in a new notebook."

notebooks(action="create", title="Sombrero")
evaluate("Plot3D[Sinc[Sqrt[x^2+y^2]], {x,-4,4}, {y,-4,4}]", target="notebook")
=> [3D surface plot rendered in the live notebook]

"Give me a Chebyshev polynomial explorer with a degree slider."

evaluate("Manipulate[Plot[ChebyshevT[n, x], {x, -1, 1}], {n, 0, 30, 1}]", target="notebook")
=> [live slider panel in the notebook; interactive code is auto-routed to the front end]

Quick start

Prerequisites: Mathematica 14.0+ (15+ recommended) with wolframscript on your PATH, and the uv package manager.

# One command, pick your client:
uvx mathematica-mcp-full setup claude-desktop   # or: cursor | vscode | codex | gemini | claude-code

Restart Mathematica (so the addon loads) and restart your editor. Then verify:

uvx mathematica-mcp-full doctor

Done - ask your agent for a plot.

The PyPI package and CLI are named mathematica-mcp-full.

Manual installation, per-client configuration details, and troubleshooting live in the Installation Guide.

The lean default

Agents see a consolidated 12-tool surface (~2.9k tokens of schema) instead of the classic 82 tools (~15k tokens) - a 5x cut in the context an agent pays before doing any work, with the same engine underneath. Prefer everything? MATHEMATICA_PROFILE=classic restores the full pre-1.0 surface, and MATHEMATICA_TOOLSETS adds opt-in extras to lean.

• Tool reference: Technical Reference - Lean Profile Tools

• Profiles and toolsets: Technical Reference - Tool Profiles

• How it stays fast: Technical Reference - Architecture

How it compares

Runs alongside the official Wolfram Local MCP (setup <client> --with-official configures both side by side). The differentiator is live notebook / front-end automation:

Who this is for

Documentation

• Installation Guide: manual setup, per-client configs, troubleshooting

• Migration Guide: upgrading between versions

• Technical Reference: architecture, tools, and configuration

• Security Model: threat model, permissions, and vulnerability reporting

• Benchmarks: performance data and reproduction steps

• Examples: worked agent conversations (symbolic calculus, notebook analysis)

• Contributing: development setup, testing, and PR process

• Changelog: version history

License

MIT License