Open integration surfaces for Layerz — structured financial modeling infrastructure.
The repos here are not the product. They are entry points: open standards, tools, plugins, and recipes that let AI agents and finance practitioners work with structured financial models — without going through the web interface.
Curated recipes for doing financial work with AI agents: prompts and patterns for FP&A, modeling and reporting, battle-tested in the field. The lowest-friction entry point: copy a prompt, use it now.
We share the recipe, not the black box. The cookbook ships prompts you can read, audit and own, never installable skills. Two kinds: working prompts that do the task in-session, and builder prompts that have Claude forge or harden your own skill, script or FINANCE.md.
Each recipe names where a prompt stops being enough, and points back to finance-md (structure) and layerz-mcp (persistence) when it does.
→ Recipes · Template · Contributing
The standard for encoding financial conventions as machine-readable files.
A FINANCE.md file placed in a financial modeling project encodes the conventions that apply to all models in that project: EBITDA definition, FX policy, fiscal year, normalization rules. Any agent that reads it knows the rules — without being told again at every session.
---
finance_md: "0.1"
model_name: "Project Granite — LBO model"
currency: EUR
glossary:
Marge brute commerciale: "Revenue minus purchase cost only. Excludes logistics."
---
## EBITDA
Excludes IFRS16 — PE sponsor convention. Operating leases stay above EBITDA
as rent expense.
## Currency & FX
Single reporting currency (EUR). Foreign subsidiaries translated at closing
rate for BS, average for P&L.The Layerz MCP server for creating and editing financial models from the terminal — a remote endpoint that works with any MCP client (Claude Code, OpenAI Codex CLI, Claude Desktop, MCP Inspector).
One connection per client. Once registered, your agent gets financial modeling expertise and the full Layerz API surface — so it can build a P&L model autonomously via the MCP tool calls.
# Claude Code
claude mcp add --transport http layerz https://app.layerz.cc/mcp
# OpenAI Codex CLI
codex mcp add layerz --url https://app.layerz.cc/mcp && codex mcp login layerzA Claude Code skill that turns a YAML + CSS spec into a self-contained, navigable HTML presentation — pitch decks, board updates, investor reports — via a zero-dependency Ruby renderer. No JS in the output, no build step.
Clone, launch Claude Code in the repo (the skill is auto-discovered), then describe your deck:
git clone https://github.com/layerzlabs/slides-for-claude.git
cd slides-for-claude && claude
# then: /layerz-slides build a Series A pitch for Acme Vision→ Skill · Templates · Examples
We publish primary research on how finance actually works, with sources and method, free to read and to cite:
- The finance engineer role — first-party corpus: 1,061 employer job boards probed, 27,278 open postings scanned, 37 qualifying requisitions coded by hand. Full method, adjudication rules and archived postings published. The role has no research literature, so we built the measurement.
- Spreadsheet errors and risks — how often spreadsheets contain errors, and what they have cost
- LLM reliability — hallucination, consistency and reasoning benchmarks
- AI in finance — adoption, use cases and measured outcomes
- Forecast accuracy and budget cycles
- Finance function cost · FP&A productivity · Fractional CFO
Writing on AI and financial modeling from the practitioner's side, including what the finance engineer job description actually asks for, Claude for financial modeling, building a model that doesn't drift and token efficiency in finance, lives on the blog.
Layerz separates the structure of a financial model (logic, DAG of variables, timelines) from its data (values, scenarios). The structure is versioned, reusable, and readable by AI agents. Excel remains a compatibility output — not the source of truth.
