Skip to content
FLORA

How do I add image and video generation to Cursor, Codex or Claude Code?

FLORA Team

Add the FLORA MCP server, https://agents.flora.ai/mcp, to your coding agent and sign in once. Claude Code takes one command, Codex takes one command plus a login, and Cursor takes a JSON entry. The agent then generates images and video with FLORA's models. Our first image from Claude Code took 30 seconds and cost $0.055.

Key takeaways

  • One server URL, https://agents.flora.ai/mcp, works in Claude Code, Codex and Cursor.
  • Claude Code: claude mcp add --transport http flora https://agents.flora.ai/mcp.
  • Codex: codex mcp add flora --url https://agents.flora.ai/mcp, then codex mcp login flora.
  • A 16:9 GPT Image 2.5 Flare image from Claude Code took 30 seconds and cost $0.055.
  • A project-scoped config file lets a whole team share the same FLORA setup through the repo.

At a glance

What it doesGives a coding agent FLORA's image, video, audio and 3D tools
Works inClaude Code, OpenAI Codex (CLI, IDE extension, desktop app) and Cursor
ModelsEvery model in FLORA, from GPT Image 2.5 to Seedance 2.5
InputsPlain-language requests, local files and repo data such as CSVs
OutputsImage and video URLs, saved on a FLORA canvas
Typical cost$0.055 per GPT Image 2.5 Flare image; about $2.48 for 5 seconds of Seedance 2.5 video
Setup timeOne command and a browser sign-in

What you need

Key features

  • Local files: Coding agents can upload images from your disk to FLORA before editing or animating them.
  • Loops over data: The agent can read a CSV or folder and send up to 20 generations per call.
  • Team config: Claude Code's .mcp.json and Codex's .codex/config.toml can be committed so teammates get the same server.
  • Quotes in dollars: Any request can be priced before it runs, which matters for video.
  • Skills: flora_discover_skills hands the agent tested workflows such as flora-batch-generate.

Best for

  • Developers building sites or apps who need real images instead of placeholders
  • Generating images from data in the repo, such as product lists
  • Animating a folder of stills in one task
  • Teams that want image generation inside the tools they already use for code

Step by step

  1. Step 1: Add the server to your agent

    Run the command for your client. In Cursor, add the JSON entry to ~/.cursor/mcp.json for yourself or .cursor/mcp.json in the repo for the team. The install guides linked above cover scopes and troubleshooting for each client.

    # Claude Code
    claude mcp add --transport http flora https://agents.flora.ai/mcp
    
    # Codex
    codex mcp add flora --url https://agents.flora.ai/mcp
    codex mcp login flora
  2. Step 2: Sign in to FLORA

    The first time the agent calls a FLORA tool, a browser tab opens. Sign in with your FLORA account and approve. In Claude Code, /mcp should list flora. In Codex, /mcp does the same.

  3. Step 3: Ask for an image where you need it

    Describe the image and say where it goes. The agent picks a model, writes the prompt, calls FLORA and can save the file into your project or drop the URL into your code.

    Use FLORA to make a 16:9 hero image for the docs landing page: an open laptop on a light wooden desk beside a paper notebook, a pencil and a ceramic mug, the code editor on screen softly out of focus, late afternoon window light, muted natural color, no readable text. Use GPT Image 2.5 Flare and save it to public/images/docs-hero.png.
  4. Step 4: Add video when the still is right

    Ask the agent to animate the approved image with a video model such as Seedance 2.5. Ask for a quote first, since video costs dollars per clip rather than cents.

    Quote animating docs-hero.png with Seedance 2.5 for 5 seconds at 720p, aspect ratio auto. If it is under $3, run it and give me the MP4 link.

What happened when a coding agent generated media through FLORA?

On October 16, 2026, we made the docs hero image above from a Claude Code session with the FLORA MCP. The agent called flora_create_generations once with model t2i-gpt-image-2-5-flare, aspect_ratio 16:9, quality high and resolution 1k, into a FLORA project, then read the result with flora_list_generations.

The image took 30 seconds and cost $0.055. It came back at 1280 x 720 pixels: a laptop on a light wooden desk with a spiral notebook, pencil, a speckled ceramic mug and a plant, a window behind with a city skyline in low sun.

What went wrong: we asked for the code on screen to be out of focus, and it came back fairly sharp, with lines of colored code visible. It is not readable at page size, but for a hero that must carry no text at all, ask for the screen to be dark or turned away. The low sun also came in warmer than the muted look we asked for; "overcast" holds the color down better than "late afternoon".

An open laptop showing a code editor on a light wooden desk with a spiral notebook, a pencil, a ceramic mug and a plant, in front of a window with a city skyline at sunset
The docs hero generated from Claude Code with GPT Image 2.5 Flare, raw output.
Client
Claude Code
Model
t2i-gpt-image-2-5-flare, 16:9, 1K
Tools called
flora_create_generations, flora_list_generations
Cost
$0.055
Time
30 seconds
FLORA MCP setup by coding agent
ClientWhere the config livesTeam sharingSign-in
Claude Code~/.claude.json, or .mcp.json in the repoCommit .mcp.json (--scope project)Browser sign-in on first tool call
Codex~/.codex/config.toml, or .codex/config.toml in the repoCommit .codex/config.toml in a trusted projectcodex mcp login flora
Cursor~/.cursor/mcp.json, or .cursor/mcp.json in the repoCommit .cursor/mcp.jsonBrowser sign-in on first tool call

Coding agent or chat client for generation: pros and cons

Pros

  • Coding agent: uploads local files and writes results straight into your project.
  • Coding agent: loops over CSVs and folders for batch jobs.
  • Chat client such as Claude on the web: no install, a custom connector and a sign-in.

Cons

  • Coding agent: a config step per client, and Codex uses TOML where Cursor uses JSON.
  • Chat client: works from public links or canvas images rather than files on your disk.
  • Both: video takes minutes, so let the agent poll while you work on something else.

What can the agent do once FLORA is connected?

It can call every FLORA tool: list models, quote and run generations, read the canvas, upload files and load FLORA skills. In practice that covers images, video, audio and 3D from any model FLORA lists.

  • Generate images for pages, posts and app screens.
  • Edit an existing image with an image-to-image model.
  • Animate a still with a video model such as Seedance 2.5 or Kling 3.0 Pro.
  • Run one image per row of a CSV, up to 20 per call.

Can the whole team use the same setup?

Yes. Install at project scope and commit the config file. In Claude Code that is .mcp.json; in Codex it is .codex/config.toml in a trusted project; in Cursor it is .cursor/mcp.json. Each teammate still signs in to FLORA once with their own account.

Do I need an API key?

No. All three clients sign in to FLORA with OAuth in the browser. To disconnect, remove the server from the client and revoke it in FLORA under Settings, then Connected apps.

Frequently asked questions

Does this work in the Codex IDE extension and desktop app?

Yes. The Codex CLI, IDE extension and desktop app share the same MCP config, so one setup covers all three.

How much does a generation cost from a coding agent?

The same as anywhere else in FLORA. Our GPT Image 2.5 Flare image cost $0.055. A 5-second 720p Seedance 2.5 clip quoted at $2.48.

Why does Claude Code show flora as pending?

Sign-in has not happened yet. Ask Claude to call any FLORA tool, such as listing your techniques, and the browser sign-in opens.