Your AI agents,
our ad legends.

Bring your design system. Keep your brand.

Import your design system straight from Claude Design into your Ad Legends brand over MCP — colors, type, and tokens. Then every ad the Legends generate stays on-brand, automatically. Export ours back out the same way. Full companion story → /use-with-claude-design · Claude connection guide →

Already creating ads with ChatGPT?

Put ChatGPT on Ad Legends for Creative Legends and Brand Memory. Pretty frames stay in ChatGPT Images 2.0. The performance tune is here. Create ads with ChatGPT → · ChatGPT connection guide →

Fast Video

Prepare a quick spot from your direction, a saved ad or a campaign. Review the source, cast and 20–60-credit quote, then create and recover the about-10–30s video through prepare_video_spot, create_video_spot and get_video_spot.

Generate ad kits

Generate Meta, LinkedIn, Google Search, static YouTube Demand Gen, TikTok carousel, X Website Card, or ChatGPT chat-card kits. Saved-creative image kits cost 2 Legend Credits per new image or card; Google Search text kits cost 3. Reuse completed outputs, keep the original, and review creative and platform requirements before publishing.

Just connected? Call get_started

The catalog is large on purpose. get_started maps your request to the next real tool: make an ad for a brand, or plan an image from your direction. Review the cost before generation.

Clone a master brand

When enabled, preview and confirm an independent product or regional brand, then follow copy progress. Selected Brand Memory, stored assets and supported external images, PDFs and fonts become independent stored copies. Zero credits; owner and unrestricted credential required. Local language and market overrides do not translate research.

Create a brand from a URL

Point the agent at any website. create_brand_from_url takes one URL in and returns Brand Memory plus a brandId — voice, colors, identity, in one call. For an explicitly requested blank or duplicate shell, create_manual_brand instead creates a separate, retry-safe manual brand without discovery or dedupe.

Import your real brand fonts

Send the Google families and complete role map together: one payload imports into private storage, assigns, locks, renders, and verifies the Casting Library card — no UI steps.

Generate Fast Ads on demand

Choose Auto, model-rendered Brand Candy, or exact-composited TrueCast — then verify the lane that actually shipped.

Create or remix any image

Describe an image or edit, attach up to 14 references, and review the free plan and exact cost. Approve generation, follow progress, then fetch the images for your next project. Brand and campaign grounding are optional; private download links can be refreshed, and saving into Brand Memory or a campaign is your choice.

Animate approved Fast Ads

Suggest or revise the motion script, render a 5–10 second video, recover durable animation history, and retry from feedback without losing the prior script or lineage.

Make a whole video ad from a brief

Hand the agent a plain-English brief and it writes the script, casts and locks the characters and product across every shot, renders the scenes, and edits them into one finished cut — poll it like any run.

Resize finished ads

Turn one approved ad into social, display, OOH, print, mail, merch, or custom pixel formats — recomposed for the surface, not just cropped.

Import existing campaigns

Hand the agent a pasted brief or extracted campaign and save it as a Legend-enhanced idea for the brand.

Lock & Load a campaign

Promote an idea or a Fast Ads concept into a first-class campaign, read it in full, set its Campaign Memory — the key line, brief, design system and cast — lock it, and every ad, post and video generated for it stays on the message.

Run resumable Idea Sessions

Find a session by brief or idea, reopen the brief-as-run deck, star/tag/pass ideas, generate more into the same lineage, preflight a fresh-eyes Jury, recover its verdicts, and hand the winner into its first campaign moves.

Organize complete creative lineages

Find or create Projects, assign one canonical growing Idea Session, preview the accessible family from strategy to brief to idea to campaign and finished work, add only what you approve, then annotate, reorder, archive, restore, or reopen the pitch.

Create private Legends

Draft a personal Creative Legend, approve and create it privately, update its profile/avatar, activate it for your roster, and share access with your team.

Let the Legends judge the work

Submit a complete Idea Session, selected ideas, or ads to the Legend Jury, then pull the feedback into revision briefs with scores, strengths, weaknesses, recommendations, and next steps.

Give your ad to a Legend

Already have an ad, or one your agent just wrote? Hand it over in one call. Say what the ad is for and set your hard rules. The rules are binding: a version that breaks one is repaired once, then flagged, and it cannot win. A Legend writes the brief, judges the ad against its job, returns three better versions with visual direction, what changed, and a status each, then re-judges all four blind with the slop measured. A featured Legend from the roster takes the ad by default, and a private Legend of yours who would also suit it comes back as a suggestion beside the take; ask for a Legend by name and you get exactly that Legend. The take names the Legend and why, and says plainly when a featured Legend stood in. No brand required. 5 Legend Credits, and reading a take back is free. A take runs about three to five minutes, reports how long each stage took, and tells you the stage it is on while it is still running; one that repairs a version can run past a five minute client timeout, and reading the take back recovers it. Want to see it rendered? One more call renders the winning version as three finished ads through Fast Ads, the exact copy set in the image and verified, 9 credits for the set.

Find your strategy

Describe what you're advertising — or just a URL — and get 3 distinct strategic angles to choose from, the same engine behind Get Started. Optionally spark unexpected angles first.

Write + refine strategic briefs

Generate, find, read, and safely edit the full brief copy, strategy metadata, audience links, and brand association before the pixels.

Plan the media from any starting point

Use a stable Ad Legends campaign, bounded context your agent already holds, or an existing brand. The five-credit run returns cited channel recommendations, exact allocation, recoverable history, and only the next actions you can really execute — it never buys or publishes media.

Atomize into social posts

Turn one piece of raw material into a run of on-brand social posts, ready for your calendar.

Create + generate audiences

Create manual ICPs immediately, or generate AI audiences with a pollable run — then query and edit them from inside your tools.

Size up every audience

Run Audience Atlas for cited TAM/SAM/SOM, stack-ranked ICPs, missing-audience discovery, live progress, history, and a safe Stop — then turn the winner into a brief, ads, or media plan.

Read + update memory

Entities, research, identity, guidelines, font specimens, and typography roles — query and edit your brand from inside your tools.

Open the team shortlist

Search shared favorites across Fast Ads, strategies, briefs, campaigns, ideas, Reimagine sessions, decks, directions, and brand assets — with creator and keeper attribution plus complete campaign child packages.

Stay ahead of credits

Agents can check Legend Credit balance before a run and return safe top-up or upgrade links when a generation would otherwise stop.

Put it on a schedule

Point a scheduled agent at the server and fetch fresh creative every morning — hands-off.

Write a Cold Open (CMO outbound)

Independently verify the exact public page before copy; search results are discovery, not proof. Thin research continues only as complete uncited emails; a verified current-role conflict parks drafting. Use the four proven defaults or explicitly mix stock, authorized personal-skill versions, and one contextual wildcard. Rank every complete plain-text email under the shared inbox and four-Legend evaluator. Long-running and replay-safe; modeled estimates only, with no compose, share, delivery, send, or HTML.

Claude

Cowork, Desktop, and Claude.ai use an account-based custom connector and OAuth — no API key. Ad Legends is not yet in Anthropic’s Connectors Directory, so add the trusted remote server manually. Claude Code uses the same endpoint through its MCP commands. The full guide covers both paths.

Step-by-step guideClaude ConnectorsCustom connector guide

ChatGPT and Codex

Use Ad Legends now as a personal plugin: ask Plugin Creator to add it in the ChatGPT desktop app, or add the production MCP URL through ChatGPT web Developer Mode. Both connect with OAuth. Codex desktop can also add the remote MCP server directly. The full guide covers every path.

Step-by-step guideBuying ads in ChatGPTChatGPT Plugins

Microsoft Copilot

Add Ad Legends as an OAuth-connected MCP tool in a Copilot Studio agent, then publish that agent to Microsoft 365 Copilot and Teams. Managed tenants may have an IT approval path before wider distribution.

Step-by-step guide

Cursor

Preferred path: install Ad Legends from the Cursor Directory, then sign in with OAuth — no API key. One-click fallback: Add to Cursor on this page. Manual fallback: add https://www.adlegends.ai/api/mcp/brands as a custom MCP connector. Ask in the editor. The agent must call get_started first.

Cursor DirectoryAdd to Cursor

Grok Bot

Connect Ad Legends from the same Cursor Directory / plugin catalog. Sign in with OAuth. Ask in chat. Same brand memory, same advertising legends, same get_started first step — the operator case for the catalog Cursor uses to build. Not an API-key CLI.

Add the connector

In VS Code or any other MCP client, add the hosted URL as a custom connector. CLIs mint a scoped key at /settings/mcp.

Sign in & approve

OAuth clients sign in and approve scopes + brands — no key. CLIs use the key you minted.

Ask in plain language

Say “make an ad for this brand/URL,” “give this ad to a Legend,” or “build a media plan from this brief.” The agent calls get_started, then follows the next tool — it should not have to guess from 169 names.

Which ad kits can my agent generate?

Through the full customer MCP, use create_ads with pack_type to request a kit from the start, or make_ad_pack to adapt an existing Fast Ad; reframe_concept remains an alias. Available kits are Meta Feed and Reels/Stories with Ads Manager copy; LinkedIn square and portrait feed with Campaign Manager copy; Google Search responsive search ad text; YouTube Demand Gen square and landscape images; TikTok three-card carousels with shared copy; X square and landscape Website Cards; and a ChatGPT square chat card with three copy options. Making a kit from saved creative costs 2 Legend Credits per new image or card: 4 for a full Meta, LinkedIn, YouTube or X kit; 6 for TikTok; 2 for ChatGPT. A Google Search copy kit costs 3 credits; Search-first creation also costs 3 credits for three concepts. Completed valid outputs are reused without another charge. Originals stay available. Review creative, destinations and platform requirements before publishing; kits do not publish ads or configure campaigns.

Can my agent clone a master brand for a product or region?

When brand cloning is enabled, an owner can use preview_brand_clone, review the selected Brand Memory families and local overrides, confirm clone_brand, then poll get_brand_clone_status. The full customer MCP requires an unrestricted credential for this journey. Copying costs zero credits and retains independently usable stored assets and supported external HTTPS images, PDFs and fonts without re-encoding. Preview and start share actor admission with web; status polling remains read-only. Language and market overrides do not translate research. Copies are independent; parent updates and AI adaptation are future work. Keep the original requestId, fingerprint and options for retries.

What is the Ad Legends MCP server?

It is a Model Context Protocol (MCP) server that lets AI agents and apps connect directly to Ad Legends — to read and write your brand memory and full brand guidelines, import a design system, and generate finished ads, all from inside the tools you already use.

Which AI tools can connect to Ad Legends?

Claude Cowork, Claude Desktop, and Claude Code; ChatGPT desktop through a personal Plugin Creator install; ChatGPT web through Developer Mode; Grok Bot; Cursor; editor agents like VS Code; and any MCP-compatible client — including open-source agents such as Hermes and OpenClaw.

Do I need to write code to connect?

No. Claude web, Desktop, and Cowork can add Ad Legends as a custom connector, then sign in with OAuth. In the ChatGPT desktop app, ask Plugin Creator to add Ad Legends as a personal plugin; on ChatGPT web, add the hosted MCP URL through Developer Mode. Claude Code, desktop Codex, and other direct MCP clients can add the hosted server themselves.

How are these ads different from generic AI ads?

Most ad generators forget the big idea. The Ad Legends creative engine was trained on the minds of the most celebrated and awarded creative directors on the planet, so connecting an agent is like hiring an elite agency — it works from your brand’s memory, voice, and design system to deliver a real idea, on brand, rather than generic blank-prompt AI output.

Can I bring my design system from Claude Design?

Yes. Import your design system from Claude Design directly into your Ad Legends brand over MCP, and every ad the Legends generate stays on-brand — your colors, type, and tokens. The full companion story — Legends, Strategist, Fast Ads, and Idea Sessions — lives at /use-with-claude-design.

Can an agent import brand fonts and set typography roles without opening the app?

Yes. Send one set_typography_roles payload with every Google fontFamilies[] entry and the complete locked assignments[]. The server preserves upright and italic instances plus variable axes in private storage, renders the multi-family role card, and returns verified inventory. For a licensed non-Google font, use upload_font_file first with its license and official/interim-fallback status, then assign the returned specimenId — with no public R2 URL or UI step.

Can an agent create or manage my private Creative Legends?

Yes. An agent can draft a private Creative Legend for your approval, create it as private-only, update its profile or avatar, activate it for your personal roster, and share or revoke access for existing Ad Legends users. Public roster publishing and featured placement stay in the app/admin flow.

Can I put ad generation on a schedule?

Yes. Point a scheduled agent — for example a Claude Cowork automation or a cron-driven script — at the MCP server, and it can fetch fresh, on-brand ads automatically on whatever cadence you choose.

Can an agent animate a Fast Ad over MCP?

Yes. For an owned Fast Ads V3 image ad, the agent can suggest or revise a motion script for free, let you edit or approve it, render a 5–10 second animation at 2 Legend Credits per second, and poll until the video is ready. get_ad_session includes durable animation history for lost-ID recovery, while feedback revisions and failed-run retries preserve the script and prior-animation lineage.

Can an agent choose how Fast Ads renders my logo?

Yes. An agent can inspect logo readiness, set the brand-wide default, or choose Auto, Brand Candy, or TrueCast for one Fast Ads V3 run. Auto routes from the current logo provenance and curation signals. Brand Candy lets the image model render the logo in-scene as free flight, with no inline logo policing. TrueCast uses the exact deterministic logo-compositing pipeline. The finished session reports both the requested choice and the lane and treatment that actually shipped.

Can an agent run Audience Atlas over MCP?

Yes. An agent can start Audience Atlas, follow its real research and scoring progress, review the full cited TAM/SAM/SOM report and stack-ranked audiences, inspect run history, or safely stop a run. A finished winner can chain directly into a strategic brief, ads, and a media plan.

Can an agent create a media plan without an Ad Legends campaign?

Yes. create_media_plan can use bounded context already held by the agent, including the objective, business or product, audience, timing, geography, constraints, and budget. A stored brand is optional enrichment, not a prerequisite, when the connected full-customer credential is unrestricted. The same tool can instead use a stable campaign ID from list_campaigns or preserve the existing brand-first request. Each intentional run costs 5 Legend Credits, retries reuse one stable requestId, list_media_plans recovers lost sessions, and get_media_plan can return a compact summary-only replay. It recommends channels and allocation; it never buys media, activates spend, or publishes a campaign.

Can an agent organize Projects and creative lineage over MCP?

Yes. An agent can find, create, edit, archive, and restore Projects; assign one canonical root Idea Session even when given a generate-more child id; preview its brief and contained ideas as explicit choices inside the accessible strategy-to-brief-to-idea-to-campaign creative family; add only the approved work items; annotate, reorder, or remove organizing links; and reopen the in-app pitch flow. Project actions preserve the underlying work and cost no Legend Credits.

My agent just connected — how does it know what to do next?

Call get_started first. That tool (plus the server’s getting-started guide) maps a plain request like “make an ad for this brand or URL” onto the next real call — create the brand from the URL if needed, check Legend Credits, then generate ads — so the agent does not have to guess from the full catalog.

Can an agent write a Cold Open (CMO outbound email) over MCP?

Yes. Call get_started with intent write_cold_open, then research_cold_open (5 Legend Credits, Creator+) for an independently fetched exact-page evidence receipt. Thin research continues only as complete uncited emails; a verified current-role conflict parks drafting. Omit candidateManifest in draft_cold_open for the four proven defaults, or call list_cold_open_archetypes and choose a bounded mix of exact stock refs, immutable authorized personal-skill versions, and the gated contextual wildcard. Repeat grade_cold_open unchanged while it checkpoints one candidate per call; its final response ranks every complete email under the same inbox, adversarial, four-Legend, safety, and tie-break policy as the web tool. Exact retries replay cached work; modeled outcomes are estimates. The server never composes, shares, delivers, sends, invents a gift, or renders HTML.

Can my agent hand an ad it already wrote to a Legend?

Yes. Call start_legend_take with the copy you have (headline, body, button, an optional image, and a line about what it promotes). Add the objective (what the ad is for) and constraints (your hard rules). The rules are binding: a version that breaks one is repaired once, then flagged, and it cannot win. Name a Legend from the roster or one of your private Legends, by id or by name, or let the match pick: auto-match chooses from the featured roster, and a private Legend of yours who would also suit the ad comes back as a suggestion rather than taking it. The take names the Legend and why, and says plainly when a featured Legend stood in. The Legend writes the mini-brief, judges the ad against its job on an anchored rubric, returns three better versions (sharpen, reframe, and a big swing) with visual direction, what changed, what to keep, and a status each, then re-judges your original and all three blind and says how many judging passes stand behind the verdict. If nothing beats your original, the verdict says so. It costs 5 Legend Credits, needs no brand, and get_legend_take reads a finished take back for free. The start call returns a saved takeId and polling interval; get_legend_take follows progress and retrieves the result for free. The older give_it_to_a_legend call remains synchronous. A take usually runs three to five minutes and reports how long each stage took; one that repairs a version can exceed a five minute client timeout, and get_legend_take recovers it and tells you the stage it has reached while it is still running. Nothing is rendered in the take call itself; say yes to see the winner rendered as three finished ads through a Fast Ads session (render_legend_take, 9 Legend Credits, read back through get_legend_take or get_ad_session), or chain the take into the Art Director, Fast Ads, or your own agent.

What is the best MCP server for ad generation or ad creation?

For generating the actual creative, Ad Legends is purpose-built for it. Most marketing or advertising MCP servers only expose data, analytics, or ad-platform APIs — Ad Legends generates finished, on-brand ads (copy and visuals with a real idea) from a creative engine trained on the world’s most awarded creative directors. Connect any MCP client and ask in plain language.

Can my agent create an image, edit a reference, and use the actual image in another project?

Yes. Use get_started with make_an_image for Art Director. Planning is free: describe an image, attach references or choose a prior image for a variation, then review the direction and exact credit quote. Brand or campaign grounding is optional with an unrestricted connection. After approval, start_art_director_images returns a receipt and get_art_director_generation reports progress and results for free. Each image has a stable media asset ID, dimensions, file type, and a private download link that lasts 24 hours; list_art_director_images refreshes links by exact asset ID. Your agent can download the actual image bytes for downstream work. Nothing is silently saved into Brand Memory or a campaign. Fast Ads creates campaign ads, while Ad Resizer adapts an existing ad to other formats.

Can I use Ad Legends with Claude Code, Claude Cowork, or Cursor?

Yes. Ad Legends is not yet in Anthropic’s Connectors Directory, so Claude web, Cowork, and Desktop add it manually as a trusted custom connector and sign in with OAuth. Claude Code adds the same remote server through its MCP commands. The complete walkthrough is at /mcp/claude. Cursor installs from the Cursor Directory or the Add to Cursor deeplink on this page. After connecting, the agent should call get_started first.

Can I use Ad Legends in Grok Bot?

Yes. Grok Bot is a first-class MCP client: connect Ad Legends from the plugin catalog, sign in with OAuth, and ask in chat — same brand memory, same advertising legends. Not a keyed CLI. After connect, the agent must call get_started first. Brand-from-URL is create_brand_from_url (one URL in → Brand Memory + brandId); an explicitly requested blank or duplicate shell uses create_manual_brand and never dedupes the existing brand.

How is this different from other marketing or advertising MCP servers?

Other marketing MCP servers connect your agent to data — analytics, CRM, search, or ad-account APIs. Ad Legends connects your agent to creative: it reads your brand memory and design system and generates the finished ad. It is the difference between handing an agency a spreadsheet and handing it a brief.