Adobe Firefly MCP Server & CLI

A free, open source Adobe Firefly MCP server and CLI that lets Claude Code, Codex, Cursor and other AI agents generate and edit images, make video and composite products.

Navid Moazzezby Navid Moazzez·Updated Oct 5, 2026·11 min read·
Your computer
  1. 1

    Add it with a single command in a terminal.

    Put your own client ID and secret in place of the examples.

    claude mcp add firefly \
    -e FIREFLY_CLIENT_ID=YOUR_FIREFLY_SERVICES_CLIENT_ID \
    -e FIREFLY_CLIENT_SECRET=YOUR_FIREFLY_SERVICES_CLIENT_SECRET \
    -- npx -y @thenavidm/firefly-mcp-cli@latest
  2. 2

    In Claude Code, type /mcp. Adobe Firefly should show as connected.

    /mcp

Needs Node 22 or newer and an Adobe Firefly Services project. Set it up.

It's free and open source. If it saves you time, star it on GitHub, so more people find it.

Rate this tool

This free Adobe Firefly MCP server and CLI gives your AI real access to Adobe Firefly Services: image generation, Image Model 5 edits, video, fills, expansions, composites and upscaling. It generates and edits images, makes short videos, fills and expands images, and composites products into scenes.

It's one install with two ways in. Claude, Codex, Cursor or any other MCP app calls its 14 tools for you, and the same tools work as a CLI that agents like Claude Code, Codex and OpenCode run, or that you type yourself.

In this guide, I'll show you:

  • What the Adobe Firefly MCP server & CLI is, and what you can ask it
  • How to install it and check that it works
  • When to use the MCP server, and when the CLI
  • Every tool, and how to keep your account safe

Every tool on this page comes from the server itself, and every cost was measured in Claude Code.

key_takeaways.mdTL;DR

Key takeaways

It gives Claude, Codex, Cursor and any other MCP app 14 tools for Adobe Firefly: images, Image Model 5 edits, video, fills, expansions, composites and upscaling.
Every tool that spends Adobe credits waits for your approval, and uploading a reference doesn't spend any.
It needs a Firefly Services project in Adobe Developer Console, which your Adobe organization provides.
The same tools run as a CLI, which costs your AI nothing until it runs.

What is the Adobe Firefly MCP server & CLI?

The Adobe Firefly MCP server & CLI is a free, open source program that lets AI agents generate and edit images and video with Adobe Firefly for you, in two ways. The MCP server is what an AI app like Claude, Codex or Cursor connects to, through MCP (Model Context Protocol), the open standard AI apps use to call outside tools.

You ask in plain language. Your AI picks the right tool, and the server makes the call to Adobe's Firefly Services API.

The CLI is the same program as commands. firefly-cli list-custom-models runs the same code your AI runs when you ask which custom models your project can use, whether an agent like Claude Code runs it or you do.

In an AI OS, the layer that connects your AI to your context, your data and your tools, a server like this is how your AI reaches Adobe Firefly. I explain how the parts fit together in how to build an AI operating system, and here's how I use AI myself.

What can you ask it?

Once it's set up, you ask the way you'd ask an assistant. These are real prompts it handles:

Try asking
Generate a square product photo of a ceramic cup in warm morning light, with Image 5.
Use this reference and change its background to a soft blue studio wall.
Fill the masked area of this photo with a small green plant.
Expand this image into a wide hero banner.
Make a five-second video of a slow camera move through a sunlit forest.
Which custom models can my Adobe project use?

The second one shows why this is useful. Your AI uploads the reference, writes the edit and runs it, and you only approve the part that spends credits.

It works well next to my Midjourney MCP server and Bannerbear MCP server, so your AI can generate images in Midjourney and fill your Bannerbear templates in the same chat. Every server I've made is on my MCP servers page.

For your prompts, my free AI image prompt generator and image to prompt generator help too.

How to install the Adobe Firefly MCP server

Pick your app in the box at the top of this page. Each one is a single command or a pasted block, and they all run through npx, so there's nothing to install first.

Windsurf, Zed and Cline take the same server, though they don't have a tab in the box yet. Follow the box's Cursor tab, and put its block here instead:

AppWhere the block goes
WindsurfIn the Cascade panel, open the Actions menu (…) and choose Open MCP config file, then paste the block.
ClineClick the MCP Servers icon in the Cline panel, open the Configure tab, click Configure MCP Servers, and paste the block.
ZedGo to Settings, then AI, then MCP Servers, and choose Add Server, then Add Local Server. Enter the block's command, arguments and variables, and Zed saves them under context_servers.
Before you start0/3

Watch out: Firefly Services is Adobe's API for businesses, separate from a Firefly subscription in the app. If your organization doesn't have it, ask your Adobe admin or representative first.

Set up your Adobe access

The server signs in with a server-to-server credential from your Firefly Services project.

Get your client ID and secret0/5

If your project uses scopes other than Adobe's defaults, set them in FIREFLY_SCOPES. To rotate the secret, make a new one in Developer Console and update your app's config.

Check that it works

Run the doctor with --network. It checks your settings, then signs in to Adobe to prove the credential works, without generating anything.

npx -y @thenavidm/firefly-mcp-cli doctor --network

When every line has a check mark, restart your app and ask which custom models your project can use. A working sign-in doesn't prove your organization pays for every model, so try one small image first.

In Claude Code, /mcp also shows whether the Adobe Firefly server is connected. It's one of the Claude Code commands worth knowing.

Use the Adobe Firefly CLI

The CLI is the same 14 tools as commands. AI agents that run commands, like Claude Code, Codex and OpenCode, use it on their own, and you can type the same commands in a terminal or a script.

Every tool name becomes a command with dashes, so generate_image5 runs as firefly-cli generate-image5.

npm install -g @thenavidm/firefly-mcp-cli
firefly-cli
firefly-cli list-custom-models
firefly-cli verify-credentials

The bare firefly-cli lists every command, firefly-cli which <words> finds the command for a task, and firefly-cli <command> --help shows what a command takes. Anything that spends Adobe credits needs --confirm, the terminal's version of your approval.

firefly-cli install <app> adds the server to Claude Code, Codex, Claude Desktop, Cursor, VS Code or Gemini CLI, in each app's own format.

These are the flags you'll use most:

FlagWhat it does
--jsonPrints JSON
--compactPrints the JSON on one line
--select a,b.cKeeps only those fields, and a dotted path goes deeper
--agentMachine mode: compact JSON, no prompts, and it never confirms anything
--confirmLets a write that asks first go ahead
--dry-runShows what a write would do, without doing it

A script can branch on the exit code:

Exit codeWhat it means
0It worked
1Something unexpected went wrong
2The command was typed wrong, or a write needed --confirm
3It wasn't found
4Adobe Firefly rejected the credentials
5Adobe Firefly failed or couldn't be reached
7You hit a rate limit, so wait and try again
10Nothing is set up yet

MCP server or CLI: which one?

Both are the same program with the same 14 tools. The difference is when your AI pays for them, in the tokens it reads.

Every message, in an app that loads every tool
MCP server
16,800 tokens
CLI
Nothing
Every message, in Claude Code
MCP server
370 tokens
CLI
Nothing
When Adobe Firefly comes up
MCP server
Nothing more, or in Claude Code the tools it picks
CLI
2,300 tokens, once
20 messages with Adobe Firefly in one, every tool loaded
MCP server
336,000 tokens
CLI
2,300 tokens
Works in Claude Desktop's chat
MCP server
Yes
CLI
No, there's no terminal there
Works in a script, a cron job or CI
MCP server
No
CLI
Yes

An app that loads every tool up front sends the full list with every message, whether you mention Adobe Firefly or not. Claude Code doesn't by default: its tool search sends only the tool names and the server's instructions, and loads a tool's full definition when your AI reaches for it.

The CLI costs nothing until Adobe Firefly comes up. Then your agent reads its skill file once and runs the commands it needs. With the skill added, Claude Code also lists its one-line description with every message, about 100 tokens.

When the whole conversation is about Adobe Firefly, the gap closes. Then the MCP server is the better experience, because you ask in plain language and your AI never has to look up a command.

How to spend less

In Claude Code, leave tool search on, which it is unless you've set ENABLE_TOOL_SEARCH=false.

Turn the server off when you aren't making images. In Claude Code that's the /mcp panel, and every AI app has its own switch. FIREFLY_READ_ONLY=1 cuts it to the three reading tools.

Or install the CLI and connect the server later, on the days it earns its place. Every tool stays in reach, and the standing cost drops to nothing.

Every number here was measured on October 5, 2026 in Claude Code 2.1.286 with Claude Opus 5.5. It's the same one-line prompt with and without the server connected, once with tool search off so every tool loads, and once with Claude Code's default. The skill was measured the same way, and other apps and models count tokens a little differently.

Image Model 5, references and jobs

Image Model 5 has its own tool, generate_image5. Ask for Image 5 and your AI uses it. To edit your own picture, it uploads the file first with upload_image, which spends nothing, then refers to it in the edit.

Video takes a while, so generate_video hands back a job right away. Your AI checks on it with get_job_status, and checking first stops it from paying for the same video twice.

Files are saved only when you ask for them, to FIREFLY_OUTPUT_DIR, which is ~/outputs/images unless you change it.

Every Adobe Firefly tool

The tables below list all 14 tools, grouped the way the repo groups them. Of these, three only read, and 11 change something or spend Adobe credits.

Images

generate_image
What it does
Generate images.
Kind
Asks first
generate_image5
What it does
Generate images with Image5.
Kind
Asks first
generate_similar
What it does
Generate similar images.
Kind
Asks first
generative_expand
What it does
Expand image.
Kind
Asks first
generative_fill
What it does
Fill image.
Kind
Asks first
upscale_image
What it does
Upscale image.
Kind
Asks first

Composites

generate_object_composite
What it does
Generate object composite.
Kind
Asks first
precise_composite
What it does
Generate precise composite.
Kind
Asks first
adaptive_composite
What it does
Generate adaptive composite.
Kind
Asks first

Video

generate_video
What it does
Generate video.
Kind
Asks first

References

upload_image
What it does
Upload a local JPEG, PNG, WebP, TIFF or JXL image, up to 15 MB.
Kind
Writes

Jobs

get_job_status
What it does
Read an existing Adobe async job by jobId.
Kind
Reads

Your account

verify_credentials
What it does
Verify OAuth credentials or a supplied access token without generating media.
Kind
Reads
list_custom_models
What it does
Read custom models available to the Adobe project.
Kind
Reads

Is the Adobe Firefly MCP server safe?

Every generation spends Adobe credits, and none come back. So 10 tools ask first: everything that makes or edits an image or a video. They're the ones marked "Asks first" in the tool tables above.

Uploading a reference doesn't ask, because it spends nothing, and reading jobs and models never changes anything.

When your AI calls one of them, the call needs a yes first. Where you use it decides who gives it:

  • Claude Code (2.1.246 and later) asks you before every call, in every one of its permission modes.
  • An app that can show forms asks you in a form.
  • Any other app takes your AI's confirm: true as the yes, so tell your AI to check with you first.
  • An agent with no person to ask needs FIREFLY_CONFIRM=model, so its confirm: true counts in every app.
  • In a terminal, you add --confirm to the command.

If the answer is no, nothing reaches Adobe Firefly.

Make it read-only

Set FIREFLY_READ_ONLY=1 and every write disappears from the tool list, leaving three reading tools. FIREFLY_ALLOW_DESTRUCTIVE=0 is the middle setting: uploads and reads still work, and nothing that spends credits does.

Keep a log of every write

Set FIREFLY_AUDIT_LOG to a file path. The server writes one line per attempted write, allowed or blocked, with what it was about to do and who approved it.

Watch out: Prompts and job details are data, not orders. A note in a job can't tell your AI to make another image, and FIREFLY_READ_ONLY=1 is the safe setting for an agent working on its own.

Your data

Your requests go straight to Adobe, with no server of ours in between and no tracking. Your client ID and secret stay in your app's config or your shell, and they're never part of a tool call.

The audit log, when you turn it on, leaves out your prompts, file paths and media links. Files you ask for stay on your computer.

Adobe Firefly MCP server settings

Every setting is an environment variable, in your app's config or your shell.

FIREFLY_CLIENT_ID
Default
none
What it does
Holds your Firefly Services client ID
FIREFLY_CLIENT_SECRET
Default
none
What it does
Holds that credential's client secret
FIREFLY_ACCESS_TOKEN
Default
none
What it does
Holds an access token you already have, instead of the secret
FIREFLY_SCOPES
Default
Adobe's defaults
What it does
Sets the scopes, if your project's differ
FIREFLY_OUTPUT_DIR
Default
~/outputs/images
What it does
Picks the folder for files you ask to save
FIREFLY_READ_ONLY
Default
off
What it does
1 hides every write
FIREFLY_ALLOW_DESTRUCTIVE
Default
on
What it does
0 refuses everything that spends credits
FIREFLY_ALLOW_SPENDING
Default
on
What it does
The 2.x name for FIREFLY_ALLOW_DESTRUCTIVE, still read when that one isn't set
FIREFLY_AUDIT_LOG
Default
none
What it does
Names a file that gets one line per attempted write
FIREFLY_SERVICES_CLIENT_ID
Default
none
What it does
Adobe's tutorial name for FIREFLY_CLIENT_ID
FIREFLY_SERVICES_CLIENT_SECRET
Default
none
What it does
Adobe's tutorial name for FIREFLY_CLIENT_SECRET
FIREFLY_SERVICES_ACCESS_TOKEN
Default
none
What it does
Adobe's tutorial name for FIREFLY_ACCESS_TOKEN
FIREFLY_USER_TOKEN
Default
none
What it does
Optional user-level token for custom models
FIREFLY_REQUEST_TIMEOUT_MS
Default
30000
What it does
Each request's deadline
FIREFLY_POLL_TIMEOUT_MS
Default
300000
What it does
How long to wait for a generation job
FIREFLY_POLL_INTERVAL_MS
Default
2000
What it does
Time between job checks
FIREFLY_SURFACE
Default
full
What it does
search lists three tools that find, describe and run the rest
FIREFLY_TOOL_TIMEOUT_MS
Default
none
What it does
Gives up on any tool after this many milliseconds
FIREFLY_CONFIRM
Default
human
What it does
model lets your AI's confirm: true approve on its own, for an agent with no person to ask
FIREFLY_HTTP_PORT
Default
8787
What it does
Sets the port for --http
FIREFLY_HTTP_HOST
Default
127.0.0.1
What it does
Sets the address for --http, and any other needs a token
FIREFLY_HTTP_TOKEN
Default
none
What it does
Sets the bearer token every --http request needs
FIREFLY_HTTP_ALLOWED_ORIGINS
Default
none
What it does
Lists other websites allowed to call --http, comma separated
FIREFLY_DEBUG
Default
off
What it does
1 prints debug lines

Troubleshooting

Run the doctor first. It names the step that failed and the fix.

What you seeWhat to do
The sign-in failsThe client ID or secret is wrong, or the credential isn't OAuth Server-to-Server. Copy both again from Developer Console
A model or a feature is refusedYour organization's Firefly Services plan doesn't include it. Ask your Adobe admin
A video seems lostIt's still running. Ask your AI to check the job before making another one
"…so it will not run without confirm: true"That's the safety check working. Approve it when your app asks

If the server doesn't show up in your app at all, run the command your app runs, in a terminal, and read the error.

More free tools for visual content

Turn a post into an image to share, or make a 3D model from a picture.

FAQs about the Adobe Firefly MCP server & CLI

Here are the questions people ask most about the Adobe Firefly MCP server and CLI.

The Adobe Firefly MCP server & CLI is a free, open source program that lets an AI app like Claude, Codex or Cursor generate and edit images and video with Adobe Firefly.

It has 14 tools for images, Image Model 5 edits, video, composites, upscaling, uploads and jobs, and the same tools run as a CLI that agents like Claude Code and Codex use, or that you type yourself.

An MCP server is a standard way to give an AI app real access to a tool, so it can act instead of guessing.

MCP stands for Model Context Protocol, and Claude, Codex, Cursor and many other AI apps speak it.

Every server I've made is on my MCP servers page.

The Adobe Firefly CLI is the same program as the Adobe Firefly MCP server, run as commands.

AI agents like Claude Code, Codex and OpenCode run them for you, and you can type them in a terminal or a script.

One npm package installs both.

Use the Adobe Firefly MCP server in an AI app with no terminal, like Claude Desktop's chat, and the CLI in a terminal, a script or a cron job.

In an AI app that can run commands, like Claude Code, Codex or Cursor, the CLI is the lighter choice, because it costs nothing until it runs.

Yes, the Adobe Firefly MCP server is free and open source under the AGPL-3.0-or-later license.

Generating spends the Adobe credits on your own Firefly Services plan, and the server adds no cost of its own.

You need Firefly Services, Adobe's API for businesses, which your Adobe organization provides.

A Firefly subscription in the Adobe apps doesn't include API access.

Yes, every image and video spends credits from your own Firefly Services plan, and none come back.

That's why each one waits for your approval first.

Yes, your AI uploads the photo, which costs nothing, then runs the edit you ask for.

Fills, expansions, Image 5 edits and composites all work on an uploaded picture.

To disconnect the Adobe Firefly MCP server, revoke or rotate the credential in Adobe Developer Console, which cuts access right away.

Then remove the server from your AI app's config.

The Adobe Firefly MCP server works with any app that speaks MCP.

That includes Claude Code, Claude Desktop, Codex, Cursor, VS Code, Windsurf, Zed, Cline, GitHub Copilot CLI, Gemini CLI, OpenCode, OpenClaw, Antigravity and Hermes.

I did.

I'm Navid Moazzez, and I made the Adobe Firefly MCP server & CLI as one of my free tools on navid.me.

You can read more about me.

The code is on GitHub under the AGPL-3.0-or-later license, and the package is on npm.

More MCP servers & CLIs

Related free tools

Navid Moazzez

AI business strategist

Navid Moazzez helps creators and founders master AI and build their own AI Operating System (AI OS) to automate their business and life.

Navid.me is reader-supported. When you buy through links on this site, I may earn an affiliate commission. Read the affiliate disclaimer.

Free AI newsletter

The most actionable AI newsletter for founders

Every week, get proven AI strategies, curated tools, and step-by-step systems to grow your audience, create better content, and build a profitable creator business.

No fluff, no filler, no BS. Just five minutes each week that might level up your online business and life.

P.S. Sign up now to get free access to my ultimate AI tools guide for creators.

Loved by 10,000+ readers
Loved by 10,000+ readers