Jacobian MCP
An MCP server with a library of typed math operations: the agent finds one, runs it and passes the result on to the next
- 320 ★ on the repository
- Last commit
Extended access
Why this level
- Runs Python computations on your system and starts through uvx or npx
- The setup command edits the configuration of the chosen agents, but shows the changes first
- Some operations use the external programs Singular and QEPCAD if you installed them
Install
- Works in
Manual install
npx jacobian@latest setupThe wizard detects agents, shows the changes and writes them only afterwards. Needs Node.js 20.17 or newer and uvx.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Runs Python computations on your system and starts through uvx or npx
- The setup command edits the configuration of the chosen agents, but shows the changes first
- Some operations use the external programs Singular and QEPCAD if you installed them
This is third-party code. Review the repository files before installing.
What it does
Jacobian gives the agent two tools. `math.find` searches for an operation by mathematical need or shows the exact contract of a chosen one, and `math.run` executes it and returns a typed result. Each operation solves one narrow task and guarantees one checkable property of the result. Exact answers are marked as exact; approximation and incompleteness are stated explicitly. The built-in library covers algebra, exact linear algebra and graphs, plus bounded SAT and SMT problems. Separate operations deal with probability, geometry and topology. Computation runs on SymPy, NetworkX, Z3 and Python-FLINT, which install with the package. The same library is available from the terminal through `jacobian inspect` and `jacobian run`, and from Python.
Who it is for. For researchers, students and engineers who tackle math and proof problems with an agent and want exact computation in place of mental arithmetic.
Good fit when
- The agent must compute exactly the determinant of a rational matrix or a gcd with Bézout coefficients
- You want to test a conjecture on small cases through SAT or SMT with a bound
- The task reduces to a graph: find a path, a coloring or check isomorphism of two graphs
- You build a solution from small checkable steps and want the agent to compose them from operations
- You need the same computations in a terminal and in a Python script
Not a fit when
- You need one large solver for a whole domain: the library offers narrow primitives and bakes in no ready strategy
- You need a stable interface: contracts of experimental operations change between releases
- You run Alpine or another musl system: the full mandatory package stack cannot install from PyPI there
Example request
Connect Jacobian, find the extended Euclid operation with math.find, then compute the gcd of 84 and 30 with Bézout coefficientsMCP
- Transport
- stdio
- Authentication
- not required
Limitations
The server is in a pre-stable 0.23.0 release, so contracts of experimental operations may change between releases. Tests cover CPython 3.12 and 3.13 on glibc Linux x86-64. Other systems may install the packages, though no tests cover them. A small subset of operations needs Singular and QEPCAD, which are not installed for you. The `npx jacobian@latest setup` command requires Node.js 20.17, 22.13 or 23.5 and later, plus uvx on PATH. Remote HTTP deployment with authentication has a separate guide.
How to disable. Remove the jacobian entry from your agent's MCP configuration. If you installed through setup, run it again and clear the agent selection. A pip install is removed with pip uninstall jacobian.
FAQ
How do I connect Jacobian to Codex or Claude Code?
Run npx jacobian@latest setup and pick the detected agents. The command shows the changes before writing. For automation there are flags such as --codex and --dry-run, and --yes works only with an explicit agent list.
How do I run the server without a global install?
With uvx --python 3.12 --from jacobian jacobian-mcp. If the client needs an npm command, npx jacobian mcp works and launches the same uvx command.
How does math.find differ from math.run?
math.find matches an operation to a mathematical need or shows the contract of a chosen one. math.run executes it with input in the right shape and returns a typed result.
Is Docker required?
Not for ordinary operations: the package and its computing libraries install through pip or uvx.
How is Jacobian different from the MATLAB Agentic Toolkit?
The MathWorks toolkit connects the agent to MATLAB. Jacobian ships its own library of narrow Python operations and needs no license for a computing environment.
Discussion
No questions or reviews yet.
Related
MCP serversA global intelligence MCP server: news, geopolitical risk, markets and infrastructure exposed as agent tools
166 skills for scientific work: bioinformatics, cheminformatics, clinical data, geospatial analysis and 100+ databases
Streamlit agent skill
developing-with-streamlit
The official Streamlit skill bundled with the library: building, styling, debugging and speeding up apps, installed with streamlit skills
Google's open-source MCP server for databases: ready tools for Postgres, MySQL, BigQuery, Spanner and more, plus custom tools in tools.yaml
Official Hugging Face skills: Hub operations via the hf CLI, datasets, model training, Spaces, evals and deployment
A CLI for fine-tuning language models from one YAML: LoRA, QLoRA, DPO and GRPO, layer streaming for 8B on a 4 GB GPU, export to GGUF
README in short
The author's hypothesis is that mathematical reasoning benefits from an executable vocabulary of narrow operations the agent searches for and composes into a solution. The library supplies checked moves, and the model decides which ones to make and when to stop. The repository has detailed docs, tests and benchmarks on conjecture problems.