How does MCP work for creative work like images and video?
MCP, the Model Context Protocol, lets an AI agent such as Claude, Cursor or Codex call outside tools. The FLORA MCP server gives those agents FLORA's image, video, audio and 3D models, its canvas and its workflow skills. You ask in plain words; the agent picks the tool, quotes the cost and runs it. One mood board image cost $0.055 in our run.
Key takeaways
- One server URL, https://agents.flora.ai/mcp, connects Claude, Claude Code, Cursor, Codex and VS Code to FLORA.
- The agent reads FLORA's live model list, so it can use any of the 237 image and 299 video model entries listed on October 16, 2026.
- flora_discover_skills listed 18 FLORA workflow skills, from batch generation to script-to-video.
- Every generation can be quoted in dollars first; our mood board image cost $0.055 and took 29 seconds.
- Results land on a FLORA canvas, so work started in a chat can be picked up by a designer later.
At a glance
| What it is | A standard way for AI agents to call tools, here FLORA's creative tools |
|---|---|
| Works in | Claude, Claude Code, Cursor, OpenAI Codex and VS Code |
| Models | Every image, video, audio and 3D model FLORA lists |
| Inputs | Plain-language requests, links, local files and spreadsheets |
| Outputs | Images, video, audio and 3D files, saved on a FLORA canvas |
| Typical cost | $0.055 per GPT Image 2.5 Flare image; quotes before every run |
| Setup time | One command or one connector, then a browser sign-in |
What you need
- Claude on the web or Desktop: the Claude install guide
- Claude Code: the Claude Code install guide
- Cursor: the Cursor install guide
- Codex: the Codex install guide
- VS Code: the VS Code install guide
The options
FLORA MCP
FLORA's MCP server at https://agents.flora.ai/mcp. It exposes generation, canvas, asset, technique, action and skill tools to any MCP client, with OAuth sign-in and dollar quotes before each run.
- Image, video, audio and 3D generation across FLORA's model list
- Read and edit FLORA canvases from the agent
- Workflow skills any agent can load by name
Claude (web and Desktop)
Add FLORA as a custom connector and switch it on in a chat. Works from public links and canvas images.
Best for: Briefs, quick visuals and edits next to the writing they are for
Claude Code
Add with claude mcp add --transport http flora https://agents.flora.ai/mcp. Reads local files and repo data.
Best for: Batches from spreadsheets and images written straight into a project
Cursor
Add the server to ~/.cursor/mcp.json or the repo's .cursor/mcp.json.
Best for: Generating images and video inside a codebase
OpenAI Codex
Add with codex mcp add flora --url https://agents.flora.ai/mcp, then codex mcp login flora. One setup covers the CLI, IDE extension and desktop app.
Best for: Teams already working in Codex
What happened when an agent made media through the FLORA MCP?
On October 16, 2026, we worked through the FLORA MCP from one Claude Code session. flora_list_tools returned the tools on the connection with a purpose and safety hints for each. flora_discover_skills listed 18 FLORA skills. flora_list_models returned 237 image and 299 video model entries, each with an estimated dollar cost and time.
For the creative run, the agent made one flora_create_generations call with model t2i-gpt-image-2-5-flare, aspect_ratio 3:2, quality high and resolution 1k, and the prompt: "Flat lay mood board for a spring linen clothing collection: fabric swatches in oat, sage and clay, kraft paper tags tied with string, a few dried grasses, on a pale plaster surface. Soft overcast light from above, muted natural color, true-to-life texture, no readable text." It polled flora_list_generations until the run finished.
The image took 29 seconds and cost $0.055, at 1248 x 832 pixels. Three linen swatches in oat, sage and clay overlap in the center, with two blank kraft tags on string and dried grasses around them. The light is soft and even and the color stays muted.
What went wrong: nothing failed. The model added a small spool of twine in one corner that we did not ask for. The same session also ran image batches, model comparisons and two Seedance 2.5 clips through the same connection, with no extra setup between them.
- Client
- Claude Code
- Tools called
- flora_list_tools, flora_discover_skills, flora_list_models, flora_create_generations, flora_list_generations
- Cost
- $0.055 for the mood board image
- Time
- 29 seconds
| Area | Example tools | What the agent can do |
|---|---|---|
| Models | flora_list_models | Read every model with its parameters, estimated cost and time |
| Generation | flora_create_generations, flora_list_generations | Quote, run and read up to 20 generations per call |
| Canvas | flora_get_canvas, flora_add_to_canvas, flora_run_canvas_nodes | Read a canvas, add and wire nodes, run them |
| Files | flora_create_asset, flora_attach_asset | Upload files and place them on a canvas |
| Techniques | flora_list_techniques, flora_run_technique | Run saved multi-step workflows on new inputs |
| Actions | flora_search_actions, flora_run_action | Crop, resize, grade and other deterministic edits that spend no credits |
| Skills | flora_discover_skills, flora_create_skill | Load FLORA's workflow skills or save your own |
| Style training | flora_list_lora_trainers, flora_train_lora | Train a LoRA style from your images and use it in generations |
Working through an agent or on the FLORA canvas: pros and cons
Pros
- Agent: you describe the result and the agent writes the prompts and picks the model.
- Agent: it can read your files, spreadsheets and repo, and loop over them.
- Canvas: you see every version side by side and branch from any of them by hand.
Cons
- Agent: you review results one message at a time unless you open the canvas.
- Agent: each client needs a one-time setup and sign-in.
- Canvas: you write and run each prompt yourself.
What is MCP, in plain words?
MCP is an open standard that lets an AI agent use tools from other services. A service publishes an MCP server that lists its tools and what each one needs. The agent reads that list and calls a tool when a request calls for it. For creative work, that means a chat or coding agent can generate and edit images and video without you leaving it.
What creative jobs does it handle well?
Jobs that mix writing, data and visuals. The agent already has the brief, the copy or the spreadsheet, so it can turn that into images and clips in the same place.
- One product image per row of a spreadsheet, up to 20 per call.
- A blog header or social image written and generated in the same chat.
- A character kept consistent across stills and video.
- A still animated into a short clip with a video model such as Seedance 2.5.
- The same prompt compared across several models in one batch.
What are FLORA skills?
Skills are written workflows an agent loads by name with flora_discover_skills. Each one tells the agent how to run a job end to end, including which tools to call, when to stop and how to report cost. FLORA listed 18 on October 16, 2026.
- flora-start-here: routes a request to the right skill.
- flora-batch-generate: one output per item in a list.
- flora-refine-loop: generate, judge against a goal, refine, stop.
- flora-pdp-deck and flora-mockup-deck: product page sets and campaign mockups.
- flora-script-to-video: script to shots to clips, with approval between stages.
- flora-image-resize and flora-video-resize: one approved asset into other aspect ratios.
Is it safe to let an agent spend on generations?
Each generation is billed to your FLORA workspace, and the agent can quote any request in dollars before it runs. Ask for quotes on anything with video. To disconnect a client, remove the server from it and revoke access in FLORA under Settings, then Connected apps.
Frequently asked questions
Do I need to code to use MCP for creative work?
No. In Claude on the web or Desktop, adding FLORA is a custom connector and a sign-in. Coding agents take one command.
Which AI clients work with the FLORA MCP?
Claude on the web and Desktop, Claude Code, Cursor, OpenAI Codex and VS Code all have install guides at developer.flora.ai.
What does it cost?
Each generation is billed at the model's price. Our GPT Image 2.5 Flare mood board image cost $0.055; a 4-second 720p Seedance 2.5 clip cost $1.996.
Where do the results go?
Into a project on your FLORA canvas, with the prompt and model that made each one. The agent also returns a link to each file.