Build with MCP

The whole Poremu API is exposed as Model Context Protocol tools over Streamable HTTP. Connect once, then say "use poremu to…".

Connect

Claude Code — one command, then sign in with Google when the browser opens:

claude mcp add --transport http poremu https://api.poremu.ai/mcp

Add --scope user to make it available in every project. Ask anything that needs Poremu — or type /mcp, pick poremu, choose Authenticate, sign in and press Allow. To disconnect: claude mcp remove poremu or Settings → Developer → Connected apps.

Claude.ai / Claude Desktop — Settings → Connectors → Add custom connector → URL https://api.poremu.ai/mcp → Connect, finish the Google login.

Cursor (.cursor/mcp.json) — Cursor shows a Needs login button after you save:

{
  "mcpServers": {
    "poremu": { "url": "https://api.poremu.ai/mcp" },
    "poremu-docs": { "url": "https://api.poremu.ai/mcp/docs" }
  }
}

ChatGPT — Settings → Connectors → Advanced → Developer mode → Create connector → MCP server URL https://api.poremu.ai/mcp → authentication OAuth → Create, then sign in.

Servers · CI · headless — any client that cannot open a browser sends an API key from Settings → Developer → API keys:

claude mcp add --transport http poremu https://api.poremu.ai/mcp --header "Authorization: Bearer pk_live_…"

{ "mcpServers": { "poremu": { "type": "http", "url": "https://api.poremu.ai/mcp", "headers": { "Authorization": "Bearer pk_live_…" } } } }

Both ways act as you: tools use your workspaces, credits, plan limits and adult-verification status. Under the hood OAuth is standard 2.1 with PKCE and dynamic registration — see Authentication.

First session

use poremu: list my products
→ use the one called Lumo
→ compare the last 7 days with the week before and tell me the bottleneck
→ fill the next two weeks of the calendar from the recommended template
→ write three Instagram post variants for the launch and save them as drafts
→ if CPA is above ₩9,000 on Meta, request a 20% budget cut (I'll approve it in the inbox)
→ build a report for last week and give me the share link

Every marketing tool takes an optional workspaceId (default: your personal / only workspace) and productId (default: the product set with use_product, else the most recent one). Call marketing_help for the suggested order.

Tools

Reads are free. Writes cost the platform credits shown; media is charged at the model's price. External actions — budgets, publishing, sending, campaign creation — are never executed by a tool: they become approval requests in the workspace inbox.

Products & strategy

Metrics

Calendar, contents & media

Resources (자산)

See Resources for the REST side.

Automations

Example: "Create an automation that every weekday at 09:00 pulls yesterday's metrics, and if spend rose more than 30% emails me the numbers." See Automations.

Approvals & reports

Connected channels

When a workspace has connected ad accounts, their tools appear too: meta_list_campaigns, meta_get_insights, google_ads_list_campaigns, google_ads_get_report, ga4_run_report, sheet_read, naver_ads_list_campaigns, naver_ads_get_report (reads) and meta_update_budget, meta_pause, google_ads_update_budget, google_ads_pause, naver_ads_update_budget, naver_ads_pause (approval requests).

Media, chat & apps

Product analysis over MCP

Over MCP the thinking happens in your client, so analysis is a two-step loop instead of one server call:

1. list_products → use_product(productId)
2. analysis_brief(urls?, notes?) — sources as text (≤ 20k chars each, up to 6), brand kit, existing strategy, recommended template scaffold, instructions and the outputSchema
3. Write the seven sections per the instructions — channel kind values from the catalog only, numbers only when a source backs them
4. save_strategy(sections, audiences?, brandKit?, facts?, summary?) — sections last written by a person are skipped unless overwriteHuman: true

"Read my product with analysis_brief, write the seven strategy sections and store them with save_strategy."

Docs server

https://api.poremu.ai/mcp/docs needs no authentication and exposes this guide as tools — list_docs, search_docs(query), get_doc(slug) — and as poremu://docs/<slug> resources. Add it next to poremu and ask "read the Poremu docs and write a Python client that creates a product, fills the calendar and listens for approval webhooks".

claude mcp add --transport http poremu-docs https://api.poremu.ai/mcp/docs

Billing

An MCP connection runs the workspace in "my subscription" mode: your client does the reasoning, Poremu charges a small platform fee per marketing tool call — reads free, analysis / strategy / content writes 1, reports 5 — plus model prices for images and videos. API calls are free up to 1,000 per day, then 1 credit per 100. The web agent (POST /v1/agent/chat) and the server-side analyser also bill LLM tokens; see Credits & pricing.

Troubleshooting

• 401 with WWW-Authenticate: Bearer resource_metadata=… — the client sent no credentials; it should start the OAuth flow. Older clients that cannot: use an API key.
• VALIDATION_ERROR: productId가 필요해요 — call list_products then use_product, or pass productId.
• PENDING from a write tool — that tool is an external action; it created an approval. Point the user to the inbox.
• ADULT_VERIFICATION_REQUIRED / NSFW_DISABLED — the account has not enabled unfiltered models; fix it in Settings on the web.
• Disconnecting — Settings → Developer → Connected apps → Disconnect, or DELETE /v1/me/connections/{id}. Revoking an API key disconnects clients using that key.