Layerz
Build, version and audit structured financial models (P&L, cash flow, balance sheet) from your agent, then export them to clean Excel with live formulas.
Documentation
MCP plugin for the AI-native spreadsheet
Connect your AI agent
to a spreadsheet it can’t break.
Layerz is an MCP server your Claude agent calls natively. Read model structure, update assumptions, compute projections, export Excel, without re-explaining context every session. Named references, so the formulas hold when the agent touches them.
Connect your agent - freePrefer the REST API
MCP-compatible · BYOA · Export to standard .xlsx
The problem with AI + finance today
Claude is powerful — the layer underneath isn't.
You use Claude. You build financial analyses in chat. The next session, you start over.
There's no persistent model structure your agent can navigate. You paste numbers, re-explain assumptions, rebuild context.
What the MCP plugin gives your agent
A model your agent can actually navigate.
Your agent navigates the model. You review the outputs. The structure persists across sessions.
Read model structure
Variable names, types, formulas, dependencies.
Query computed values
Revenue, EBITDA, cash flow, any output at any period.
Update assumptions
Change a growth rate or headcount number, all outputs recompute automatically.
Export Excel
Trigger a clean .xlsx from your agent session.
BYOA — Bring your own agent
Your prompts, your data, your environment.
Layerz does not capture your Claude session. Connect your own Claude Max subscription or API key. Layerz is the structured backend, not the middleware.
Supported: Claude (MCP), any MCP-compatible agent, Claude Code.
How to connect
Three steps to a connected agent.
Add to Claude in one click
Opens claude.ai with the Layerz connector pre-filled. You confirm, sign in, done. Or set it up manually below.
1
Add the Layerz MCP server
Add the connector to your Claude configuration.
2
Authenticate
Sign in via OAuth or paste your Layerz API key.
3
Call layerz_* tools
Your agent now has access to layerz_read, layerz_update_branch, layerz_get_model and more.
→ Full MCP reference and llms.txt
MCP clients
Install once in your agent of choice.
| Client | Auth | Install |
|---|---|---|
| Claude | OAuth | Settings → Connectors → Add custom connector |
| Claude Cowork | OAuth | claude.ai → Settings → Connectors → Add custom connector |
| Claude Code | OAuth | claude mcp add -t http layerz https://layerz.cc/mcp |
| ChatGPT | OAuth | Settings → Apps & Connectors → Add new connector |
| OpenAI Codex | OAuth | codex mcp add layerz --url https://layerz.cc/mcp |
| Mistral Le Chat | OAuth | Settings → Connectors → Add custom MCP connector |
| OpenCode | OAuth | opencode.json: {"mcp":{"layerz":{"type":"remote","url":"https://layerz.cc/mcp","oauth":{}}}} then: opencode mcp auth layerz |
Full install guide on GitHub
What you can build
From prompt to model to .xlsx.
Auto-populated budget models
Your agent fills a model structure from a prompt — assumptions first, then projections. Change one variable, all outputs update. No re-run.
Persistent unit economics
Define CAC, LTV, payback period once. Query across Claude sessions without rebuilding context.
Multi-scenario analysis
Run base / upside / downside on a shared model structure. Export any scenario to .xlsx — your colleagues get a clean file, they never see the agent.
Example prompts
Say it in plain English. Your agent does the modeling.
“Show me the revenue assumptions in my latest Layerz model and explain how they drive EBITDA.”
layerz_list_models layerz_read
“Raise the churn assumption to 2.5% from March and show me the impact on year-end cash.”
layerz_patch layerz_read
“Create an upside scenario with 20% faster hiring, compare it to base, then export it to Excel.”
layerz_update_branch layerz_export
Questions, or stuck on setup? Write to [email protected].
Build the structure once. Query it from every session.
Connect your agent - free