Locofy

Designs come in from Figma, Penpot and Adobe XD, with Figma reached through the customer's account so files need not be shared by hand, and leave as React, React Native, Next.js, Gatsby, Vue, Angular, Flutter, HTML and CSS, SwiftUI or Jetpack Compose code.

That code reaches developers through a zip download, a VS Code extension, the CLI's pull and merge into a local codebase, or GitHub and GitHub Enterprise sync on Enterprise plans, and component libraries from Storybook, Material UI, Ant Design, Chakra, Bootstrap or Vuetify can be mapped in. The agent does not call business systems or outside APIs while it works.

Sourcelocofy.ai/docs/url/export-and-deployment/sync-with-githubread 2026-10-18

Locofy Lightning runs one conversion pass over a design, optimizing assets, tagging elements, setting responsiveness and detecting components, and Agent Mode then applies prompts across the generated project, merging its changes into the full project with an automated AI merge when a person applies them.

In the CLI, built in skills merge generated code into an existing codebase during pull and integrate, and Interactive Mode takes follow up instructions file by file. Frames convert one at a time on Hobby and Pro and several at once on Enterprise, and longer multistep work is left to the customer's own coding agent through MCP.

Sourcelocofy.ai/docs/plugin/agent-moderead 2026-10-18

Large Design Models read frame structure, layers, auto layout, constraints and components so the code follows the design itself rather than a flat image. Teams can push their own component library with locofy push and map Figma component properties to code props, including complex mappings through mapFn, so output reuses existing components, though custom component mapping needs Enterprise or a legacy annual plan. Screenshots, existing code files and Markdown specifications or acceptance criteria can be dropped into the CLI as extra context, and Agent Mode accepts reference images.

Sourcelocofy.ai/docs/mcp/design-system/overviewread 2026-10-18

Switching to Agent Mode first asks for confirmation that a new version of the code will be created, all Agent Mode conversations work on one working folder, and nothing reaches the full project until a person selects Apply Changes, with undo for recent updates along the way. For MCP updates to an existing codebase, teams are advised to commit first so a design update can be reviewed or reverted, and code export stays a deliberate step. Approvals per action are not offered beyond the Enterprise permission on who may export code.

Sourcelocofy.ai/docs/url/editing-code/agent-moderead 2026-10-18

ISO 27001:2022 certification and SOC 2 Type 2 compliance back the platform, and SAML single sign on is supported. Enterprise adds user role management that decides who can consume tokens and export generated code, team project sharing with access control and full data privacy. Pro content is never used to train or improve the platform while Hobby content may be, and MCP personal access tokens can be set to expire and regenerated, which deactivates the old one at once.

Sourcelocofy.ai/enterpriseread 2026-10-18

A Plan and Usage page lists every LDMtoken charge and credit by date and action, such as a Lightning run or an Agent Mode prompt, with the remaining balance and billing history. Enterprise adds visibility into each team's usage, token consumption and which members exported code. The live preview in Builder updates as code changes, but there is no step by step log of what the agent did inside a prompt and no export to monitoring tools.

Sourcelocofy.ai/docs/cli/quickstartread 2026-10-18

Skills installed into a project hold naming conventions, styling rules, shared component detection and framework patterns, the agent reads them in every session, and teams can edit them or commit custom skills to version control so the same standards apply across the team. Agent Mode keeps one working version across conversations until changes are applied. Beyond those files and the project itself, the agent keeps no memory of past sessions or preferences.

Sourcelocofy.ai/docs/cli/skills/using-skillsread 2026-10-18

Three deployment models are offered, a shared SaaS cloud, a dedicated private cloud, or self hosted on premises for enterprises, with SAML single sign on across the options. Generated code is standard and owned by the customer, so the application keeps running without Locofy, and it reaches the customer's environment as a zip, through the VS Code extension, the CLI, MCP or GitHub, including GitHub Enterprise on a company's own host.

Sourcelocofy.ai/enterpriseread 2026-10-18

Prebuilt Agent Mode prompts cover theming, the Angular Material UI kit, Bootstrap, internationalization, form validation, accessibility, animations, hover effects, filter and search, responsive tables, carousels, popups, side drawers, mobile menus, tabs, accordions, notification dropdowns, image enhancement and code quality. The CLI installs built in skills tuned to the detected framework, Locofy Components and support for Material UI, Ant Design, Chakra, Bootstrap and Vuetify speed up tagging, and full build tutorials walk through complete projects.

Sourcelocofy.ai/docs/plugin/agent-moderead 2026-10-18

Work starts from the Locofy plugin inside Figma in Design or Dev Mode, from a Figma URL pasted into Locofy Builder in the browser, from the locofy convert command in a terminal, or from a prompt such as generate code using locofy mcp for a screen in Claude Code, Cursor, Copilot, Windsurf or Gemini CLI. Each run begins with a person or the coding agent they are driving; there are no schedules, design change triggers or chat channels that start a conversion on their own.

Sourcelocofy.ai/docs/mcp/quickstartread 2026-10-18

Conversion runs on Locofy's own Large Design Models, and Agent Mode layers large language models on top for edits, with no choice of model or provider and no way to bring a key. Customers who want a particular model use it through their own coding agent, which calls Locofy over MCP while Locofy keeps doing the design conversion itself. Because Agent Mode brings in language models, its output may no longer match the Figma layer structure exactly, which a confirmation spells out before the switch.

Sourcelocofy.ai/docs/plugin/agent-moderead 2026-10-18

The Locofy MCP server, configured per project with a personal access token and project ID, gives Claude Code, Cursor, GitHub Copilot, Windsurf, Gemini CLI and other MCP clients Locofy's design aware conversion, generating code from a Figma URL and merging it into an existing codebase or updating a single section when the design changes. The Locofy CLI, signed in to the account, runs convert, pull, push, init, integrate and skills install from the terminal, and a VS Code extension fetches projects, components and styles.

Sourcelocofy.ai/docs/mcp/installing-mcpread 2026-10-18

Agent Mode can be prompted to clean up code quality, fix responsiveness or add accessibility, and the guidance is to resize the live preview to tablet and mobile widths and check each section before exporting. The Builder preview stays in sync with the code so each change can be seen at once. Locofy does not write or run tests, and checks of behavior beyond the interface are left to the customer's own tools or coding agent.

Sourcelocofy.ai/docs/url/editing-code/agent-moderead 2026-10-18

Design files arrive through the Figma, Penpot and Adobe XD connections and the generated code appears in a preview, but no agent operates a browser, a desktop or outside software, and the Builder preview is for a person to inspect. Browser work against the generated app happens only in whatever coding agent the customer connects. Figma access opens the builder's own browser only to sign in and allow the connection.

Sourcelocofy.ai/docs/plugin/quickstartread 2026-10-18