Claude Fable 5 is Anthropic's new flagship — the first model in the Claude 5 family and the first of a new "Mythos-class" tier that sits above Claude Opus in capability. For marketers, the headline isn't the benchmark scores. It's that Fable 5 is the first model reliable enough to run an entire ad production pipeline — market analysis, persona, script, scene images, video render, quality-control retry — as one long agentic session without a human unsticking it every third step. Connected to the UGC Copilot MCP server, that session ends with finished, platform-ready video ads.
This guide covers what Fable 5 actually is, why its agentic reliability matters more than its writing quality, and the exact setup and workflow for driving UGC Copilot's 13 MCP tools from Claude Desktop or Claude Code.
What is Claude Fable 5?
Claude Fable 5 is Anthropic's most intelligent generally available model, released as part of the new Claude 5 family in mid-2026. It introduces a new model tier — Mythos-class — positioned above the Opus line that topped Anthropic's lineup through 2025. Fable 5 shares its underlying model with Claude Mythos 5 (a restricted-access variant for approved organizations); Fable is the version anyone can use, with additional safety measures around dual-use capabilities.
Three things distinguish it in practice:
- Long-horizon agentic reliability. Earlier models could call tools; Fable 5 can plan and execute 20+ tool calls in sequence — with branching, error recovery, and budget awareness — without losing the thread. That's the exact shape of an ad production pipeline.
- State-of-the-art coding and reasoning. It currently leads public coding benchmarks, which matters for marketers indirectly: the same capability that writes correct code also follows 15-constraint creative briefs without dropping requirements.
- It's still Claude. The conversational, "unscripted-sounding" script voice that gave earlier Claude models the edge in our Claude vs ChatGPT scripting comparison carries through — hooks still read like a creator caught mid-thought, not ad copy.
Anthropic also shipped Claude Sonnet 5 alongside it (June 30, 2026) — a much cheaper workhorse at $2 per million input tokens and $10 per million output tokens on introductory pricing. Keep that name in mind; it changes the cost math below.
Why frontier agentic capability changes ad production
A language model — any language model — cannot render video. What it can do is orchestrate the tools that render video. The question that separates a demo from a production workflow is: how many steps can the model chain before it needs rescuing?
A real UGC ad pipeline is roughly: analyze the market → generate a script with scene-level visual prompts → generate a scene image per scene → kick off an async video render per scene → poll each render to completion → check the QC verdict → resubmit any failed render → apply text overlays → collect the final URLs. For a 3-scene ad that's 15–20 tool calls, several of them asynchronous, at least one of them conditional on a quality verdict.
With 2025-era models, agents running this pipeline stalled in predictable places: they forgot to poll a pending render, re-submitted a paid render instead of using the free QC retry, or silently dropped the "9:16 vertical" constraint by scene three. Fable 5 is the first model where we watch full pipelines complete unattended as the norm rather than the exception. That's the practical meaning of "Mythos-class" for a marketing team.
Setup: connect Fable 5 to UGC Copilot in 5 minutes
The UGC Copilot MCP server works with any MCP-compatible client — Claude Desktop, Claude Code, Cursor, Cline, Zed. For Claude Desktop, add this to your MCP configuration:
{
"mcpServers": {
"ugc-copilot": {
"command": "npx",
"args": ["-y", "@ugccopilot/mcp@latest"],
"env": { "UGC_COPILOT_API_KEY": "ugc_live_..." }
}
}
}The API key comes from your profile page (any paid plan or pay-as-you-go credit pack unlocks API access — no subscription required). Four of the 13 tools — trend analysis, hook generation, persona previews, and script previews — work without a key on a free daily quota, so you can test the connection before spending anything.
Full install instructions, including Claude Code and Cursor variants, are in our Claude Desktop MCP tutorial and the engineer's Claude Code guide. There's also a dedicated walkthrough on the Claude Desktop video ads page.
The workflow: one prompt, thirteen tools
Here's a session we run weekly, verbatim:
"Analyze the skincare market, pick the strongest trend, write a 3-scene TikTok script for [product], generate a scene image for each scene, render all three on Sora 2 standard at 8 seconds, wait for them to finish, and give me the video URLs. If any render fails QC with a free retry available, use it."
Fable 5 decomposes that into the tool sequence itself:
| Step | MCP tool | Credits |
|---|---|---|
| Market analysis | analyze_market |
1 |
| Full viral script (hooks, scenes, CTAs) | generate_script |
1 |
| Scene images (×3) | generate_image |
3 |
| Async renders (×3, Sora 2 standard, 8s) | render_video |
54 |
| Poll to completion | wait_for_video / check_video_status |
0 |
| Final MP4 URLs | fetch_video |
0 |
Total: 59 credits for a three-scene ad — about $5.90 on a pay-as-you-go 500-credit pack, or the market analysis and script for a rounding error and the renders as the real cost. If a completed render fails the automated quality check, the status response includes a retryAvailable flag, and resubmitting the identical render is free — Fable 5 reads that hint and uses it without being told twice. If you already have a script, parse_own_script converts it into render-ready scenes for 1 credit, and apply_text_overlay burns captions onto finished videos for 1 credit per call.
Which video engine should the agent pick?
Fable 5 chooses sensibly when you describe the goal, but it helps to know the menu. UGC Copilot routes renders across five engines; the agent passes the engine as a parameter on render_video:
| Engine | Best for | Credits (std / HQ) |
|---|---|---|
| Sora 2 | Actor performance, talking-head UGC | 18 / 65 (8s) |
| Veo 3.1 | Prompt adherence, multi-scene narrative | 40 / 130 (fixed) |
| Kling 3.0 | Image-to-video motion, 4K output | 32 / 50 (6.4s), 4K tier available |
| Seedance 2.0 | Motion/dance content, budget renders | 18 / 35 (4s) |
| Gemini Omni Flash (preview) | Fast 720p drafts with native audio | 40 (8s) |
The deeper comparison lives in our 3-way engine test. A good default instruction for the agent: "draft on Seedance standard, final on Sora 2 standard, upgrade the winner to HQ."
Fable 5 or Sonnet 5? (The cost question)
Honest answer: most UGC pipelines don't need Fable 5 for every run. Once the workflow is proven — the prompt is stable, the tool sequence is known — Claude Sonnet 5 executes it at a fraction of the token cost, and at $2/$10 per million tokens the orchestration cost of a full pipeline run is pennies. Where Fable 5 earns its premium:
- Designing the workflow. First-time pipeline construction, ambiguous briefs, "figure out why scene 2 keeps failing QC" debugging.
- Batch orchestration at scale. The 500-ads-per-month pattern involves budget gates, deduplication, and performance-data feedback — long-horizon work where Fable 5's reliability compounds.
- Creative judgment calls. Scoring 20 generated scripts against a brand voice and picking the 3 worth rendering.
The pattern we recommend: Fable 5 designs, Sonnet 5 runs. Build and debug the pipeline in a Fable 5 session, save the working prompt, then execute the weekly batches on Sonnet 5. Same MCP server, same tools, same credit costs on the UGC Copilot side either way.
Getting started
If you're new to the agentic side entirely, start with our plain-English MCP explainer, then the AI agent automation use-case page for the workflow shape. When you're ready: create an account, grab a credit pack, generate an API key, and paste the config block above into Claude Desktop. The free-tier tools mean you can watch Fable 5 drive the pipeline before you commit a single credit to a render.