POPJAM Logo
Servidor MCP remoto · HTTP en streaming

Servidor MCP de POPJAM:
un motor de marketing
para tu agente de IA

Una sola URL da a Claude, Cursor, ChatGPT o cualquier cliente MCP el motor de campañas de POPJAM: investigación de marca, descubrimiento de audiencias, creatividades publicitarias y simulación con personas, en 28 herramientas seleccionadas que tu agente llama directamente.

https://api.popjam.io/mcp/
conectado · POPJAM · 14 herramientasSesión de ejemplo

tú  "Crea anuncios para acme-coffee.com y dime qué audiencia gana."

→ start_campaign(website_url="https://acme-coffee.com")

   tarea 7f3a programada · lab creado · ~40 créditos estimados

→ wait_for_task("7f3a")

   RUNNING · marca y catálogo resueltos: Acme Coffee, 12 productos

   RUNNING · 2 audiencias definidas · 20 personas simuladas

   RUNNING · 4 creatividades de imagen generadas · reacciones simuladas

   COMPLETED en 6 min 12 s

→ get_campaign_results(lab_session_id)

   insight principal: "Ritual matutino" × Profesionales ocupados, interacción 82/100

agente Los Profesionales ocupados responden mejor (82/100). El enfoque del ritual superó al del descuento en todos los paneles. Yo escalaría ese.

¿Qué es el servidor MCP de POPJAM?

Model Context Protocol es el estándar abierto que permite a los copilotos de AI usar herramientas externas. El servidor MCP de POPJAM pone todo nuestro motor de marketing bajo ese estándar: tu agente puede investigar un mercado, definir audiencias, redactar conceptos de anuncios, lanzar todo el embudo de campaña y leer reacciones simuladas de personas, todo desde la misma conversación en la que haces todo lo demás.

Todo llega a tu espacio de trabajo de POPJAM. Una campaña que tu agente inicia a través de MCP aparece como una sesión de Lab en la aplicación web (misma marca, mismas audiencias, mismas creatividades, mismos resultados), para que tu equipo pueda retomar en el navegador exactamente donde lo dejó tu agente.

Veintiocho herramientas seleccionadas, no un espejo de la API

Diseñado específicamente para LLM: un pequeño conjunto de operaciones autocontenidas con contratos claros, porque las herramientas seleccionadas rinden mejor que decenas de endpoints convertidos automáticamente.

Drafting

Stateless, return in seconds

  • research_topic

    Quick web research with live sources.

  • fetch_website

    Scrape a landing page or competitor URL to clean text.

  • define_audiences

    Turn a campaign brief into structured target audiences.

  • generate_ad_concepts

    Draft on-platform ad creatives (headline, body, CTA) from a brief.

Background work

Return a task handle immediately, run for minutes

  • start_campaign

    The flagship: from one URL, run the entire funnel (brand & catalog, market research, audiences, persona panels, image creatives, simulated reactions, improved variants), persisted to a lab in your workspace.

  • start_deep_research

    Slow, multi-source deep research as a background task.

Task tracking

Follow background work without blocking a turn

  • wait_for_task

    Long-poll a task for up to 50 seconds and get its progress log.

  • get_task

    Snapshot one task: status, progress narrative, result.

  • list_tasks

    See what's running now or what finished while you were away.

  • cancel_task

    Stop a running task; persisted work stays in place.

Workspace & active lab & copilot

Read results, resume labs, converse with memory

  • get_campaign_results

    Audiences, ads, and insights from a lab; works mid-run with partial results.

  • list_labs

    Your lab sessions (campaign workspaces), newest first.

  • get_lab

    One lab's working state, audiences, and recent tasks.

  • get_active_lab

    Read the lab session you're currently working in (your selected lab).

  • set_active_lab

    Select a lab session as your active workspace; subsequent generate tools use it by default.

  • copilot

    Talk to POPJAM's in-product agent in your active lab. It knows the lab's brand, ads, and results, keeps durable conversation memory shared with the web app, narrates its work as live progress, and can trigger any generation as a background task.

Media generation

Synchronous, runs in your active lab, streams progress, persists the result. Spends credits.

  • save_ad_concept

    Persist an ad concept (from generate_ad_concepts) as a Content row in your active lab so you can generate its media.

  • generate_image

    Generate an image ad creative for a saved Content row (imgen agent). Streams progress; writes media_url back to the row.

  • generate_video

    Generate a video ad creative for a saved Content row (videogen agent, omni-flash). Streams progress; writes media_url back to the row.

  • generate_animation

    Generate an animation ad creative for a saved Content row (animategen agent, rendered with Remotion). Streams progress; writes media_url back to the row.

  • generate_carousel

    Render one 1:1 image per card for a saved CAROUSEL row (carouselgen agent): credits per card, atomic set, card 1 becomes media_url and every card_image_url is listed in order.

Managed campaigns

Draft, propose and review managed Meta campaigns — spend consent always happens in the app

  • draft_managed_campaign

    Save a managed Meta campaign draft — reuses an explicit lab or creates its preparation lab. No media generation, no paid ads.

  • list_managed_campaigns

    Your organization's managed Meta campaigns, paged and filterable by brand or lab.

  • get_managed_campaign

    One campaign's results, blockers, settings and current permissions.

  • propose_managed_campaign_settings

    Save proposed settings without confirming them or increasing spend; the owner reviews and consents on the app's campaign screen.

  • prepare_managed_campaign

    Generate a campaign's creatives on explicit request with the exact approved credit estimate — never launches paid ads.

  • propose_managed_audience_test

    Propose a narrower age-audience test within approved countries and budget, reusing approved ads; the owner reviews it before anything runs.

  • pause_managed_campaign

    Request a pause through the canonical service; returns a task handle to follow with wait_for_task.

El trabajo de varios minutos nunca bloquea un turno

Una campaña completa tarda entre 10 y 15 minutos. Nada lento se ejecuta dentro de la solicitud, porque el trabajo pesado devuelve un identificador de tarea y tu agente sigue siempre el mismo patrón de tres llamadas.

1
start_campaign(website_url)

Launch and get a handle

Returns in seconds with a task_id, the lab it will write to, and the credit estimate. The funnel (brand research, audiences, persona panels, creatives, simulations, variants) runs in the background on our infrastructure.

2
wait_for_task(task_id)

Follow the progress log

Long-polls up to 50 seconds per call and returns a human-readable narrative of what the run is doing. A campaign typically takes 10–15 minutes, so your agent checks in a few times, and it can keep doing other work in between.

3
get_campaign_results(lab_session_id)

Read results, even mid-run

Audiences, ads, and simulation insights, readable at any time. While the campaign is still running you get whatever exists so far; when it finishes, you get the full picture: the same data your team sees in the web app.

Conéctate en menos de un minuto

Pega una URL, inicia sesión con tu cuenta de POPJAM en el navegador y listo.

Claude Code

Un comando en tu terminal y, en el primer uso, el inicio de sesión con OAuth se abre en tu navegador.

claude mcp add --transport http popjam https://api.popjam.io/mcp/

Claude Desktop y claude.ai

Configuración → Conectores → Añadir conector personalizado; luego pega la URL del servidor.

https://api.popjam.io/mcp/

Cursor, VS Code y otros

Añade la URL como servidor MCP remoto en la configuración MCP de tu cliente y autentícate.

https://api.popjam.io/mcp/

El mismo espacio de trabajo, los mismos créditos, una puerta más

El servidor MCP no es un producto aparte: es otra puerta de entrada a tu organización de POPJAM. Los labs que crea tu agente aparecen en la aplicación web para todo tu equipo; la herramienta de copiloto comparte su memoria de conversación con el chat del producto; los créditos salen del mismo saldo, y la estimación se devuelve en el mismo momento en que se inicia una ejecución. Regístrate una sola vez y cada interfaz, navegador, agente, o ambos, trabajará sobre las mismas campañas.

Preguntas frecuentes

What is the POPJAM MCP server?

A remote Model Context Protocol server at https://api.popjam.io/mcp/ that exposes POPJAM's marketing engine to AI copilots. Instead of mirroring our REST API, it offers 28 curated tools: quick drafting (research, audiences, ad concepts), the full campaign funnel as a background run, task tracking, media generation (image, video, animation, carousel) with live progress, a copilot with durable per-lab memory, and managed Meta campaigns your agent can draft and review — spend consent always happens in the app. Your agent calls them like any other tool.

Which AI clients can connect?

Any MCP client that supports remote servers over Streamable HTTP: Claude Code, Claude Desktop and claude.ai (as a custom connector), ChatGPT, Cursor, VS Code, Windsurf, Cline, Goose, and others. You paste one URL, authenticate in the browser, and the tools appear.

Do I need a POPJAM account?

Yes. The server operates on your organization's workspace: labs, audiences, ads, and simulation results are persisted there and shared with the web app. Connecting signs you in with your POPJAM account; the free tier includes 500 credits with no credit card.

Does using the MCP server cost anything?

The server itself is free. Campaign runs spend POPJAM credits exactly like the web app. start_campaign returns the credit estimate in its immediate response, and a run can be stopped at any point with cancel_task (work already persisted stays in place). Quick drafting tools like research_topic and define_audiences are free to call.

Won't a 15-minute campaign time out my agent?

No, because campaign work never runs inside a request. start_campaign returns a task handle in seconds; your agent follows up with wait_for_task, which long-polls up to 50 seconds per call (safely under client tool timeouts) and returns a human-readable progress log. get_campaign_results works at any time, returning partial results while the run is still going.

How is this different from POPJAM's agent skills?

The agent skills (github.com/popjam-io/skills) are open-source instruction sets that run POPJAM-style pipelines locally inside your agent, writing to files. The MCP server runs POPJAM's hosted engine: real media generation, persisted lab sessions your team sees in the web app, credits, and a copilot that remembers every conversation. Skills are the run-it-yourself edition, while the MCP server is the product with a plug on it.

Sigue explorando

POPJAM for Slack

The team-facing sibling: chat with the copilot in threads and get campaign results pushed back where you asked.

Agent skills for Claude Code

The run-it-locally edition: free, MIT-licensed skills that run POPJAM-style pipelines in your agent.

Test an ad creative now

Upload one creative and see how AI buyer personas react: score plus reasoning, free.

¿Listo para probar tus anuncios antes de invertir?

Genera, valida y lanza creatividades publicitarias en un solo flujo de trabajo. Empieza gratis con 500 créditos.