This document is written for security, IT, and AI-enablement teams evaluating FigmaLint for use on managed/work machines. It describes exactly what the plugin does, what data leaves the machine, where it goes, and how to constrain it to an approved provider (e.g. Azure OpenAI / Azure AI Foundry).

FigmaLint is open source (ISC license). Everything below can be verified in the source.

A Figma plugin that audits components for design-system compliance, accessibility, and developer readiness, and generates handoff documentation. Analysis is performed by an AI model that you supply an API key for. FigmaLint is a client-side tool — there is no FigmaLint-operated backend that receives your design data.

• Runs entirely inside Figma's plugin sandbox: a sandboxed main thread plus a sandboxed iframe UI. It has no filesystem, OS, or network access beyond what Figma grants.

• Network access is allowlisted in the plugin manifest and enforced by Figma. Any request to a domain not on the list is blocked by the Figma runtime. The complete allowlist is:

  Domain	Purpose	When contacted
  api.anthropic.com	Anthropic (Claude) API	Only if you select Anthropic
  api.openai.com	OpenAI API	Only if you select OpenAI without a custom endpoint
  generativelanguage.googleapis.com	Google (Gemini) API	Only if you select Google
  *.openai.azure.com, *.services.ai.azure.com, *.cognitiveservices.azure.com, *.inference.ai.azure.com	Azure OpenAI / Azure AI Foundry	Only if you configure a custom Azure endpoint
  design-systems-mcp.southleft-llc.workers.dev	FigmaLint design-systems knowledge service	Best-practice lookups during analysis and the chat feature

  There is no analytics, telemetry, crash-reporting, or tracking endpoint — the list above is exhaustive.

• Keys are stored locally in Figma's per-user clientStorage, one entry per provider. They are never uploaded to any FigmaLint-controlled server.
• A key is only ever transmitted to that provider's own API, in the request auth header (Authorization: Bearer …, or api-key: … for Azure).
• Keys can be cleared at any time from the plugin UI.

1. To your selected AI provider (your key, your account): when you run an analysis or chat, FigmaLint sends a prompt containing the selected component's structure — layer names, variant/property definitions, detected tokens and hard-coded values, and accessibility-relevant attributes. This is the design context needed to audit the component. It goes only to the provider you chose, authenticated with your key. With the Azure option below, this goes only to your Azure tenant.
2. To the design-systems knowledge service (design-systems-mcp.southleft-llc.workers.dev): best-practice lookups during analysis and any questions you type into the chat. The payload is the query text plus high-level component context (component family and names). It does not include your API keys.
3. Local only: your most recent analysis result is cached in Figma's plugin storage so it can be restored when you reopen the plugin. It is not transmitted.

No fixes are written to your file without an explicit confirmation step in the UI.

API usage is billed to the account that issues the key, entirely separate from any Claude/ChatGPT/Gemini consumer or team subscription. If you use the Azure option, usage is billed to your Azure subscription, under your organization's existing cloud agreement — not to any external OpenAI or Anthropic account.

If your organization's process only approves models hosted in your own Azure tenant, configure FigmaLint's OpenAI provider to use a custom Azure endpoint:

1. Provider: OpenAI (GPT).
2. Expand Custom endpoint (Azure / OpenAI-compatible).
3. Endpoint URL — your Azure deployment's chat-completions URL. The recommended form is Azure's OpenAI-compatible v1 API: https://YOUR-RESOURCE.openai.azure.com/openai/v1/chat/completions (Azure AI Foundry endpoints under *.services.ai.azure.com also work.)
4. Deployment / model name — the name of your Azure deployment.
5. API key — your Azure resource key.

With this set:

• All model traffic goes to *.azure.com (your tenant). Nothing routes through api.openai.com.
• The Azure key is sent via the api-key header (handled automatically).
• The key is not required to be an sk-… OpenAI key; Azure/gateway key formats are accepted.

This lets an AI-enablement team deploy an approved model into Azure AI Foundry and point the plugin at it, keeping both the data path and billing inside the organization's Azure boundary.

From manifest.json: currentuser (identify the current user), teamlibrary (read published library variables so tokens can live in a separate file), and documentAccess: dynamic-page. No access to other files, accounts, or org resources beyond the open file.

• Open source, client-side only, no vendor backend receiving design data.
• Egress is manifest-allowlisted and Figma-enforced; the full list is above; no telemetry.
• Keys are local, sent only to the chosen provider.
• Can be constrained to your own Azure tenant for both data path and billing.

Questions or a deeper review: https://github.com/southleft/figmalint/issues