Fundi
Fundi (Swahili: fixer, mechanic, technician) is the agent that helps you
set up Mzizi inside your project. It is not a separate product, not a
service, not an outside actor — it is an AI agent that ships inside the
mzizi-sdk package and runs against your own codebase under your control.
Where Fundi lives
Section titled “Where Fundi lives”Fundi lives in the nyuchi/fundi monorepo
as the agent runtime of mzizi-sdk. The monorepo has three packages that
work together:
| Package | Role |
|---|---|
mzizi-sdk |
The SDK consumers install. Carries Fundi, the agent that explores and sets up Mzizi in your project. |
mzizi-mcp |
The live MCP server. Exposes Mzizi’s primitives, brand components, tokens, and architecture to any MCP-aware agent. |
mzizi-skills |
The published skills package. Encodes Mzizi’s guidelines — token rules, ecosystem-node placement, harness wiring, brand conventions — as agent-readable skills. |
You install mzizi-sdk in your project. Fundi comes with it.
What Fundi reads
Section titled “What Fundi reads”Fundi grounds every decision in two canonical sources:
mzizi-skills— the published skill bundle. This is where Mzizi’s doctrine is written down in a form an agent can execute against. Decision rules, placement guidance, naming conventions, harness contracts, token usage — all of it.mzizi-mcp— the live MCP server. This is where Mzizi’s current state lives: which primitives exist, which brand components are available, what tokens are published, what versions are stable. Fundi queries the MCP at runtime so the answers are always fresh.
Skills tell Fundi how to think about Mzizi. The MCP tells Fundi what is true right now. Together they let Fundi reason about your project against the real, current design system without you having to memorise any of it.
What Fundi does
Section titled “What Fundi does”Fundi’s job is to take a consumer project from “I want to use Mzizi” to “Mzizi is wired in correctly.” The loop is explore → plan → apply.
- Explore. Fundi inspects your repo to see what is already in place — which tokens are wired, whether the harness is mounted, which primitives you have copied in, what brand components exist, where pages live, what framework you are on.
- Plan. Fundi compares your project against
mzizi-skillsand the livemzizi-mcpcatalogue, then produces a plan of what is missing or misaligned — missing tokens, primitives that need to be installed, harness wiring that is absent, brand components you should pull in, conventions that aren’t being followed. - Apply. Fundi executes the plan against your codebase. It adds the tokens, installs the primitives, wires the harness, scaffolds the brand components, and leaves your repo in a state where the next thing you build can follow Mzizi correctly.
Humans stay in the loop the whole time. Fundi proposes; you accept, edit, or reject. There is no headless self-healing flow, no central issue tracker, no “Fundi did something to your project while you were asleep.” Fundi runs when you run the SDK.
User-facing surface
Section titled “User-facing surface”Today Fundi is consumed as part of mzizi-sdk. You import it, you call it,
you get an interactive agent that operates on your project.
A thin CLI wrapper is planned so you can run Fundi from the terminal without
writing SDK code (npx mzizi setup, npx mzizi check, etc.). The CLI will
be a wrapper around the same SDK agent — same skills, same MCP, same
behaviour, just a different entry point.
Ecosystem placement
Section titled “Ecosystem placement”In the ecosystem-node model, Fundi is conceptually node N9 — the node that helps the system get set up correctly inside a consumer project. The product is the SDK agent described above; the node is the architectural slot it occupies.
Fundi is Nyuchi-operated (the nyuchi/fundi monorepo) and Bundu-governed
(the skills and conventions it reads are Mzizi-defined, which Bundu owns as
canonical doctrine).