FormLab
Login Get Pro

MODEL CONTEXT PROTOCOL

Connect your formulation data to Claude, with MCP.

FormLab ships a Model Context Protocol (MCP) server: your formulas, batches, ingredients and test data become readable by Claude and any MCP client, 41 read tools, no copy-paste, nothing uploaded to us on the free path. Free reads a local export; Pro connects Claude to your live cloud workspace with a dedicated read-only token, always current.

Try FormLab, free How to connect

Just want to ask questions inside FormLab? The built-in AI assistant needs no setup, MCP is for driving your data from your own AI tools.

FormLab formula library - the structured data Claude reads through the FormLab MCP server

Stop pasting CSVs into chat. Let the AI query your data.

The Model Context Protocol is an open standard that lets an AI assistant read structured data through tools rather than pasted text. Instead of dumping a spreadsheet into a prompt and hoping the model keeps it straight, the assistant calls real functions against your actual database, and the answers are grounded in your data, not a lossy copy.

FormLab's MCP server exposes your formulation library that way. Ask Claude to find every formula using an ingredient, compare two recipes, surface recent test failures, or pull a DOE coverage matrix, and it queries your records directly. On the free path it reads a local export and sends nothing to us (what a tool returns goes to your AI provider, like anything you paste into a chat); on Pro it reads your live cloud workspace, read-only and scoped to your workspace.

Ask Claude things like…

41 read tools across your whole library.

Formulas

List & get formulations, compare two side by side, find the most similar to a target, and substructure-search ingredients by SMARTS.

Ingredients, inventory & documents New

List and get ingredients with CAS, supplier, density, cost and class, plus current inventory levels, lot records (with their CoA status), suppliers, and the document register (SDS, CoA, expiry), so the AI sees what's on hand and what paperwork is missing.

Batches & samples New

List & get batches and samples with their lineage, so the AI can trace material → batch → sample end to end. Compare repeat runs of one formula for reproducibility, and roll production up by formula, status or month.

Tests, DOE & coverage

List & get test results, surface recent failures, and pull a DOE matrix, coverage matrix or design-space map to reason over your experiments. Read stability trends too - drift, control chart and shelf-life, by condition.

Methods, panels & presets New

List & get test methods, test panels and step presets, the reusable procedures behind every measurement, so the AI knows how a result was produced.

Projects, notebook & equipment New

List & get projects, read lab-notebook entries, and list & get equipment, so the AI has the context and instruments around the data. Roll a whole project up by formula, status or month.

Recently expanded from 23 to 41 tools - projects, lab notebook, equipment, the test methods, panels & step presets behind every measurement, your saved DOE design records, stability trends and the design-space map, batch and project reproducibility and production roll-ups, and now suppliers and the document register are all readable too.

Every tool is read-only, the AI can query and reason over your data, but never change it.

Wire it up two ways.

Free - point Claude at a static export

# Export your data (Settings → Data → Export JSON)
# then point any MCP client at it via npx
npx formlab-mcp ~/Downloads/formlab-export.json

Local stdio; the server itself makes no network calls. Hot-reloads when you re-export. 41 read-only tools; the server code is MIT-licensed. No data yet? Try it on the demo dataset (431 KB).

In MCP Registry → View on npm →

Works with Claude Desktop · Claude Code · VS Code · any MCP-compliant client

Try it in five seconds.

Open FormLab, build a few formulas, export, and point Claude at the file. No signup: with Node.js and an MCP client such as Claude Desktop, it is one command.

Try FormLab, free See it in the tour

FormLab MCP questions, answered.

What is the FormLab MCP server?

It's a Model Context Protocol server (the package formlab-mcp) that exposes your FormLab data to AI assistants as read tools, 41 of them, spanning formulas, ingredients, inventory, batches, samples, test results, DOE, test methods, projects, notebook, equipment and lots. An MCP client like Claude calls those tools to query your real data instead of working from pasted text.

Is it free?

Yes. The file-mode server is free: point it at a JSON export and any MCP client can read your data locally. On Pro, the same tools read your live cloud workspace instead, so answers always reflect your latest data with no manual export.

Does my data get uploaded anywhere?

Not to us on the free file path: the server runs locally over stdio, reads a JSON export on your machine, and makes no network calls of its own. But whatever a tool returns goes to the model your MCP client uses (for Claude Desktop, that is Anthropic), the same as anything you paste into a chat, so only connect data you are comfortable sending to that provider. On Pro the server reads your own live cloud workspace through a read-only database role, scoped to your workspace by row-level security.

Which AI clients work with it?

Any MCP-compliant client, including Claude Desktop, Claude Code, and VS Code. It's listed in the official Model Context Protocol Registry and published on npm.

Can the AI change or delete my formulas?

No. Every tool is read-only, the assistant can list, fetch, compare, search, and analyze your records, but it cannot create, edit, or delete anything.

Where do I get it?

It's on npm as formlab-mcp, in the MCP Registry; the source and README ship in the npm package. The server code is MIT-licensed; it bundles FormLab’s calculation engine under a license that allows its use only as part of formlab-mcp (see the package’s LICENSE). Run npx formlab-mcp <your-export.json>, or try it first on the demo dataset.