Create an API token
Create a LlamaGen API token, then store it as LLAMAGEN_API_TOKEN. Never paste the real token into a chat or commit it to Git.
Create API token
Hosted Comic MCP · Streamable HTTP
LlamaGen MCP for comic agents.
Plan a complete comic in Claude, ChatGPT, Codex, Cursor, or your own agent. LlamaGen handles creation, job state, and editable visual assets through one public MCP endpoint.
4 primary comic tools + usage aliasBearer API tokenApproval before creation
Connect once
A professional setup path for every host.
Choose your product, scope, and exact configuration. Each path separates what works today from what still needs OAuth or release-version validation.
Setup availablehttps://llamagen.ai/api/mcp
Connect from Claude Code
Give Claude Code access to LlamaGen's hosted Comic MCP with one remote HTTP registration.
Add the hosted MCP
Register the endpoint at user scope and send the token from your environment in the Authorization header.
bashClaude Code CLI
export LLAMAGEN_API_TOKEN="<YOUR_LLAMA_GEN_API_TOKEN>"
claude mcp add --scope user --transport http \
--header "Authorization: Bearer ${LLAMAGEN_API_TOKEN}" \
llamagen https://llamagen.ai/api/mcpVerify, approve, create
Confirm the four primary LlamaGen tools plus the get_api_usage compatibility alias, check Comic API usage, approve the page-and-panel plan, then create and poll the generation.
textVerify connection
/mcpThe command follows Claude Code's current option ordering. Complete a real client smoke test before showing a Verified badge.
Conversational creation
See the conversation. Trust the result.
Switch between six professional comic workflows. Each preview shows the brief, the MCP calls, the approval checkpoint, and the visual asset returned to the agent.
Codex
LlamaGen MCP connected
Script to production-ready pages
Turn my 2,150-word sci-fi screenplay into a 12-page comic. Keep Mara, Sol, and the station uniforms consistent. Show me the page-and-panel plan before generating.
LlamaGen
I found the current Comic API workflow and checked the available credits. I’ll preserve the three-character bible while mapping the script into 48 panels.
- 12 pages with four panels per page
- Recurring character and wardrobe anchors
- Explicit approval before the first write
Plan approved. Generate the first delivery.
search_llamagen_docs
Comic generation contract loaded
get_comic_api_usage
Credits checked before creation
create_comic_generation
12-page job accepted
get_comic_generation_status
Generation completed
The first delivery is ready: 12 pages and 48 panels. I kept the page plan, task state, output URLs, and delivery record together for review.
Result
Verified delivery
12-page screenplay adaptation
- Pages
- 12
- Panels
- 48
- First delivery
- 68.8%
Verified delivery case with retained inputs, outputs, revisions, metrics, and SHA-256 evidence.
Inspect the evidenceThe conversation is an interface simulation. Result panels use existing LlamaGen project or retained evidence assets and are labeled accordingly.
Human-in-the-loop workflow
From brief to editable comic, with approval in the middle.
The agent reads first, reports usage, proposes a structured plan, and waits. Only then does it use the write tool. The resulting generation ID keeps status and assets traceable.
- 01
Brief the agent
Describe the story, audience, format, visual direction, and recurring characters in your usual AI workspace.
- 02
Read before writing
The agent searches current LlamaGen docs and checks Comic API credits before proposing work.
- 03
Approve the plan
Review the page-and-panel plan, continuity notes, style, panel count, and expected credit use.
- 04
Create and poll
After approval, the agent starts one generation and tracks its ID until assets or a clear error return.
- 05
Refine visually
Open the result in LlamaGen to adjust panels, layout, speech bubbles, continuity, and final delivery.
Live server contract
Four primary tools. One complete comic lifecycle.
The surface stays deliberately small: search the current product knowledge, check credits, create once after approval, then poll the same task to completion. The legacy get_api_usage name remains as an equivalent read-only alias.
https://llamagen.ai/api/mcp
1
ReadPublic docs
search_llamagen_docs
Find current Comic API, MCP, authentication, rate-limit, webhook, and error-handling guidance.
2
ReadBearer token
get_comic_api_usage
Read plan state and remaining Comic API credits before generation.
3
WriteBearer token
create_comic_generation
Create a comic task from a prompt with optional style, 1–40 panels, output size, and a public character reference image.
4
ReadBearer token
get_comic_generation_status
Poll a generation ID and inspect status, panels, assets, and errors at generation, page, or panel scope.
Verified delivery evidence
The output is the proof.
These are controlled LlamaGen delivery evaluations with the input, full-resolution output, first-delivery result, revisions, evidence hash, and known limitations preserved for review.
Exact settled provider cost and active human review time were not separately measured for these runs. Each full case discloses its review method and remaining defects.
Case 01Aug 11, 2026
Script → complete comic
How a 2,000-word script became a 12-page comic
One original 2,150-word screenplay became a complete 12-page comic with 48 panels and 6 recurring characters.
- Words in
- 2,150
- Pages out
- 12
- First delivery
- 68.8%
16 recorded revision events across 15 panels; complete PDF and CBZ retained.
SHA-256 111e16d6f5ba2b978e2b8460906bc81e3de74acafa3977a3b1a69693c3c58fc4
View full evidence
Case 02Aug 11, 2026
First delivery → targeted revision
AI first delivery, requested changes, and final delivery
A controlled change request targeted page 7, panel 2. The workflow redrew that panel, preserved the authored dialogue band, and kept every other page unchanged.
- Kept unchanged
- 47/48
- Panel redraw
- 1
- Outside-lock pixels
- 0
97.9% delivered-version acceptance after one requested modification.
SHA-256 0ff2e3cef04a9239a476aa19e84371eda58ae6055eb84392628714235a6f09a7
View full evidence
Case 03Aug 11, 2026
Multi-character continuity
Eight recurring characters across 40 comic panels
8 registered characters made 80 appearances across 40 panels. Every appearance was reviewed against disclosed identity anchors.
- Characters
- 8
- Panels
- 40
- First delivery
- 92.5%
3 targeted redraws; final 40/40 panels accepted.
SHA-256 98c7d95e7fe7663b99d3ae8f482ad6583c281c7a8a836d68d40e8c819ab1d636
View full evidenceHonest capability boundary
Conversation controls. LlamaGen creates and refines.
MCP removes handoffs for planning and task creation. It does not pretend a chat window replaces the visual studio.
Your agent plans
Claude, ChatGPT, Codex, Cursor, or your own agent understands the request, asks questions, and chooses when to call a tool.
LlamaGen generates
Comic generation, optional character references, panel assets, usage, and asynchronous job state stay inside LlamaGen.
The editor refines
Creators keep visual control over composition, bubbles, panel-level replacement, polish, and final export.
Reusable operating prompt
Give every agent the same safe workflow.
This prompt makes research, credit reporting, approval, task creation, polling, and editor handoff explicit.
agent-instructions.txtApproval-aware
Use the connected LlamaGen Comic MCP.
1. Search the current LlamaGen creation docs.
2. Check Comic API usage and report LlamaGen credits separately from this host's model usage.
3. Turn my brief into a page-and-panel plan, including character continuity notes.
4. Wait for my approval before using the write tool.
5. Create the comic generation only after approval.
6. Poll the generation status and report the generation ID, completed assets, and any errors.
7. Tell me which parts should be refined in the LlamaGen editor.MCP FAQ
Clear answers before you connect.
Product support, authentication, credits, tools, references, and completion behavior—without hiding release gates.
Turn your next conversation into an editable comic.
Create a token, choose the setup for your agent, and keep the first write behind an explicit approval step.