# Maths, CS & AI Compendium: A 20-Chapter Open Textbook With an MCP Server

> HenryNdubuaku/maths-cs-ai-compendium is an Apache-2.0 textbook on maths, computing and AI, plus a local MCP server that turns it into a knowledge base for coding assistants. Here is what the repository actually contains and where it stops.

**HenryNdubuaku/maths-cs-ai-compendium** — Become a cracked AI/ML researcher/engineer with this unconventional textbook covering maths, computing, and ML with intuition.

- Repository: https://github.com/HenryNdubuaku/maths-cs-ai-compendium
- Stars: 7,563 · Forks: 933
- Language: TypeScript
- License: Apache-2.0
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/henryndubuaku-maths-cs-ai-compendium

## What the compendium solves, and who it is for

Most maths and ML books assume prior knowledge, lean on notation, and go stale quickly. The README states that this project is an "open, unconventional textbook covering maths, computing, and artificial intelligence from the ground up", written for "curious practitioners looking to deeply understand the stuff, not just survive an exam/interview". The intended reader is an engineer or researcher who already writes code and wants the underlying maths and systems knowledge in one reading path, not a student shopping for a semester syllabus.

The scope is unusually wide for one repository. The outline runs from chapter 01 on vectors through matrices, calculus, statistics and probability, then machine learning, computational linguistics, computer vision, audio and speech, multimodal learning, autonomous systems, graph neural networks, computing and OS, data structures and algorithms, production software engineering, SIMD and GPU programming, AI inference, and ML systems design. Chapters 19 and 20 (applied AI and bleeding edge AI) appear as top-level directories but are absent from the README outline table, so their content is not described in the README. That gap is worth knowing before you plan a reading schedule around them.

## How the repository is organised: markdown chapters, mkdocs, and an MCP layer

There is no build step required to read it. Each chapter is a directory of markdown files, and the repository root carries mkdocs.yml, javascripts/, stylesheets/ and images/, which is the standard layout for a Material for MkDocs site. The README points readers at the hosted version at henryndubuaku.github.io/maths-cs-ai-compendium, so the intended flow is either read the rendered site or open the markdown directly on GitHub.

The interesting piece is the MCP server. The README says the repository "includes an MCP server that lets any AI assistant (Claude Code, Cursor, VS Code, etc.) use the compendium as a knowledge base" and that it "requires a local clone of the repo". Two constraints follow from that sentence. First, this is not a hosted endpoint you can point a tool at; the assistant and the repository have to live on the same machine. Second, because the server reads the local markdown, the quality of an answer is bounded by the quality of the chapter files in your clone. The README also mentions that the server "comes with tools for educational purposes and example implementations", but does not enumerate the tool names or their schemas, so you have to read the mcp/ directory to learn the actual interface.

A root-level llms.txt file is also present. That is a convention for giving language models a plain-text summary of a site, which fits the same goal as the MCP server: make the compendium consumable by an assistant rather than only by a human reader.

## Installing it and running the MCP server for the first time

For plain reading, there is nothing to install. Clone the repository and open the chapter folders, or use the hosted site linked from the README.

```bash
git clone https://github.com/HenryNdubuaku/maths-cs-ai-compendium.git
cd maths-cs-ai-compendium
ls
```

The listing should show the chapter directories, images/, javascripts/, llms.txt, mcp/, mkdocs.yml and stylesheets/. If mcp/ is missing, you cloned a fork or a partial checkout, and the server described in the README will not be there.

The README states that the MCP server requires a local clone, so the clone above is the prerequisite for every assistant you connect. The README does not give a package name, install command, port or environment variable for the server. Before inventing one, read the files inside mcp/ and follow whatever setup they specify.

To preview the documentation site the way the hosted version is built, the mkdocs.yml at the root is the configuration entry point:

```bash
python -m pip install mkdocs
mkdocs serve
```

The README does not document which MkDocs theme or plugins the configuration expects, so if mkdocs.yml references a theme you have not installed, the build will fail and the error will name the missing package. Treat that error message, not this article, as the source of truth for the plugin list.

## The content is uneven by design, and the outline admits it

The outline table carries a Status column, and every chapter listed there is marked Available. That is a useful signal, and it is also the only status signal the repository provides. There are no releases, no version tags and no changelog, so "Available" is a snapshot of the author's own tracking table rather than a reviewed edition.

The deeper limitation is one of form. A 20-chapter compendium spanning linear algebra, operating systems, CUDA and ML systems design is a large surface for a single author to keep correct. The README's own framing supports that: it describes the project as unconventional and intuition-first, which is exactly the register in which hand-waving is easiest to miss. Nothing in the repository indicates external review, errata tracking or a citation process. If you need to cite a definition or a derivation in a paper, this is the wrong source; there is no edition number, no DOI and no per-chapter authorship.

The MCP server has a related failure mode. It answers from whatever is in your clone. If a chapter is thin, the assistant will produce a confident answer built on thin material, and nothing in the described setup flags low-confidence retrieval. For exam preparation or interview drilling, that is a risk worth naming: verify any answer the assistant gives against the markdown file it should have come from.

## Compared with a conventional textbook and with a docs site

The obvious alternative is a standard textbook, and the difference is structural rather than qualitative. A published textbook gives you a fixed edition, an index, exercises with known answers and a citation target. This repository gives you markdown files you can read, fork and edit, an online version, and an assistant-facing layer. The trade is currency and accessibility against stability and review. The README argues for the repository's side of that trade directly, claiming that textbooks "quickly get outdated in fast-moving fields like AI".

A second comparison is with documentation tooling itself. Material for MkDocs, which the mkdocs.yml and stylesheets/ layout point to, is a general static-site generator for technical documentation, not a curriculum. It gives you navigation, search and theming, and it is the reason the hosted site exists at all. It does not give you the chapters. The compendium is the content; MkDocs is the delivery. Confusing the two leads people to expect a maintained product with release notes, when what is actually here is a book repository that happens to render as a site.

## Maintenance, licence and what upgrading actually costs

The last push to the default branch was on 2026-07-18. There are no releases, so there is no version to upgrade to and no migration path to plan. Updating means pulling main and re-reading the chapters that changed; your own notes, forks and annotations are the only local state at risk.

The licence is Apache-2.0, which permits commercial and private use, modification and redistribution, and requires that you keep the licence and notice files and state significant changes. It also includes an explicit patent grant, which matters if you plan to reuse diagrams or text inside a product. This is a description of the licence text, not legal advice; if you intend to redistribute the material commercially, read LICENSE at the repository root and take your own counsel.

One practical upgrade cost is easy to overlook: because the MCP server reads a local clone, any assistant you have wired to it is answering from the revision you cloned. Pulling a new revision changes what the assistant says. If you keep a local clone for a course or a study group, pin the commit you reviewed rather than tracking main, so that a chapter rewrite does not silently change the answers your group receives.

## Conclusion

Adopt it if you want a breadth-first reading path across maths, systems and modern ML, and if you are willing to read 20 chapter folders on GitHub rather than a typeset PDF. Skip it if you need a citable, peer-reviewed reference with a stable edition, or if you need a hosted knowledge base, since the MCP server requires a local clone. Before relying on it, check the status column in the README outline for the chapter you need, open that chapter's markdown file to confirm it is written rather than stubbed, and read the mcp/ directory to see which tools the server exposes.

## FAQ

### What does maths AI stand for in the Maths, CS & AI Compendium?

In this project the pairing refers to the two halves of the title: the mathematical foundations (vectors, matrices, calculus, statistics, probability) and artificial intelligence (machine learning, NLP, computer vision, audio, multimodal learning and the rest). The README describes it as a textbook covering maths, computing and AI from the ground up.

### Is a CS degree heavy on math, and does this compendium cover that math?

A CS degree curriculum is not discussed in the repository, but the mathematical ground usually associated with one is covered: chapters 01 to 05 handle vectors, matrices, calculus, statistics and probability, and chapter 13 covers discrete maths alongside computer architecture and operating systems.

### Is AI and ML maths hard, according to the Maths, CS & AI Compendium?

The README does not rate difficulty. It states the project exists because most textbooks "bury good ideas under dense notation, skip the intuition, assume you already know half the material", and positions the compendium as an intuition-first alternative for practitioners rather than a gentle introduction.

### What type of math is used in AI, and which chapters cover it?

The outline lists linear algebra (chapters 01 and 02), calculus and optimisation (chapter 03), statistics (chapter 04) and probability with information theory (chapter 05) before the machine learning chapter. Later chapters add graph theory in chapter 12 and discrete maths in chapter 13.

## Sources

- [HenryNdubuaku/maths-cs-ai-compendium on GitHub](https://github.com/HenryNdubuaku/maths-cs-ai-compendium)
- [Issues](https://github.com/HenryNdubuaku/maths-cs-ai-compendium/issues)
- [License: Apache-2.0](https://github.com/HenryNdubuaku/maths-cs-ai-compendium/blob/main/LICENSE)
- [README](https://github.com/HenryNdubuaku/maths-cs-ai-compendium/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/henryndubuaku-maths-cs-ai-compendium