choose-your-plungeai-door

Picks the right PlungeAI entry point before you write anything — the MCP server (https://mcp.plungeai.com/v1) for live agent operation, the One API (https://api.plungeai.com) for code you write or generate, the Ocean CLI for a terminal or CI workflow, or Ocean Studio for a human building visually — and names the setup or capability skill to load next. Use when the user asks which PlungeAI door, API, or interface to use, says "how do I use PlungeAI", "what's the difference between the MCP server and the One API", asks "does PlungeAI have a CLI/SDK/UI", or is choosing between MCP, REST, CLI, and Studio before writing any code. For the mechanics once a door is picked, use plungeai-mcp-setup (MCP), plungeai-api-setup (REST/codegen), plungeai-sdk-setup (official SDKs), or plungeai-cli-setup (terminal); for wiring a specific coding tool's agent to PlungeAI use plungeai-in-<tool>; the full capability map lives in plungeai-platform.

Download zip (opens in a new tab) · View raw SKILL.md (opens in a new tab)

PlungeAI (internally "Ocean") is a Cloudflare-native agent runtime: every agent, tool, and workflow runs as a deployed edge service. There is no local mode — everything you execute runs on the deployed platform, and everything you read (catalogs, contracts, results) is live. You reach it through one of four doors, all backed by the same identity and the same data (a workflow saved through one door appears in every other door and in Studio).

The four doors

The One API has three levels — all stay

REST now covers bots, and the same key opens all three levels; none is deprecated in favour of another (plungeai-api-setup):

1. Compatible wire — /v1/chat/completions, /v1/models and the OpenRouter-shaped /api/v1/models: an OpenAI or OpenRouter client works unchanged (models only).
2. Standard REST — one route per resource, JSON or YAML in and out: saved workflows, agents and bots (/v1/workflows, /v1/workflows/{id}/permissions, workspace files, inbox, chat bindings), /v1/bot-groups, /v1/channels, /v1/executions (continue, steer, files), /v1/schedules (cron or @once) with triggers (/v1/triggers/webhook, email, hooks), /v1/connections, /v1/memory/recall, /v1/skills, /v1/templates. Start here.
3. Agent Call door — POST /v1/execute: the body is the program (a bare task, a workflow:, or agent: <saved id>@n + input); always through the engine, so a saved bot keeps its stored policy.

The same bot verbs exist as actions on the MCP tools (plungeai_workflow permissions, files, bindings and groups; plungeai_executions steer and cancel; plungeai_schedule wake and triggers) and as ocean-cli commands — pick the door by who is calling, not by what it can do.

Rules of thumb

• Operating live from a chat/agent context → MCP. One tool call per intent; the platform renders final user-ready markdown — relay it verbatim. Setup: plungeai-mcp-setup. Full guide: https://mcp.plungeai.com/docs#0-overview (opens in a new tab)
• Generating code for a user's project → One API. Point codegen at GET /v1/openapi.json; never invent routes. Setup: plungeai-api-setup. Building or running a bot: author it with plungeai-bot-agent, then drive it over the verbs above.
• An SDK in your own code (TypeScript, Python, Go, Rust) → One API through its official SDK. Setup: plungeai-sdk-setup.
• A human or CI script driving a terminal → Ocean CLI. Setup: plungeai-cli-setup.
• Anything a door refuses (an OAuth consent, visual editing) → a person in a browser: Ocean Studio, or the hosted connect link POST /v1/connections/{provider}/link mints. An API-key connection goes over REST: POST /v1/connections/{provider}.
• Connecting a specific coding tool's agent (Cursor, VS Code, Claude Code, Codex, Windsurf, Replit, Gemini CLI, OpenCode, Lovable, Bolt, v0, …) rather than deciding which door in the abstract → jump straight to that tool's plungeai-in-<tool> skill; it tells you which door that tool uses and gives the exact config.
• Same identity behind every door: one ozk_ key on MCP, the One API and the CLI (ocean-cli login); an optional Studio session adds the CLI's account verbs (keys, balance, usage, teams).

Get a key first

Self-service ozk_ keys: Dashboard → One API → Keys (https://dashboard.plungeai.com) → create → copy it once (shown only at creation). Ask the account owner only if you need a shared team key minted under someone else's account. The same key works on MCP, the One API and the Ocean CLI (ocean-cli login); the CLI also takes an optional Studio session for its account verbs (keys, balance, usage, teams, secrets).

Which capability for which job

Once you know your door, plungeai-platform is the full capability map. Quick router:

• One-shot capability call ("search the web", "convert this PDF") → plungeai-agents (prompt-driven) or plungeai-tools-connectors (structured, typed params).
• Multi-step pipeline ("research A, B, C in parallel, then synthesize") → CNL workflow, author it with plungeai-workflows.
• Open-ended goal needing judgment ("investigate X, use whatever tools you need") → one bounded harness mission, plungeai-missions.
• "Every morning / every hour" anything → plungeai-scheduling.
• Long-running lead-gen / outreach style batches → plungeai-campaigns.
• Instruction packs injected into a run (skills, plugins, experts, personas) → plungeai-skills-plugins.
• A standing agent with its own policy, memory and chat → a bot, authored with plungeai-bot-agent (several of them: /v1/bot-groups).
• "Remember this" / long-term memory → plungeai-memory.
• Raw LLM inference in your own code (chat/embeddings, model fallback) → plungeai-models.
• "What can PlungeAI do" / finding an id → plungeai-discovery.
• "Why did that run fail / how long / what did it cost" → plungeai-results-traces.

Discovery first (every door, non-negotiable)

Catalogs are live — the agent registry, tool contracts, model list, and workflow inventory change without notice. Never assert what exists from memory:

• MCP: plungeai_list_agents {search: "<capability>"}
• One API: GET /v1/discovery/search?q=<capability>, GET /v1/agents, the live contract at GET /v1/openapi.json
• CLI: ocean-cli discovery search <query> / ocean-cli agents get <id>

Verify you're in

• MCP: call plungeai_whoami — identity card (user id, tier, key label, rate window).
• One API: curl -s https://api.plungeai.com/health (no auth) proves the router is up; curl -s https://api.plungeai.com/v1/whoami -H "Authorization: Bearer ozk_YOUR_KEY" proves the key ({user_id, via, tier, key_label}).
• CLI: ocean-cli doctor (exit 0 = ready).

Related skills

• plungeai-mcp-setup — connect and operate the MCP server.
• plungeai-api-setup — auth, first call, OpenAPI, codegen for the One API.
• plungeai-cli-setup — install, auth, full command reference for ocean-cli.
• plungeai-sdk-setup — the official One API SDKs (TypeScript, Python, Go, Rust): install, key, first call.
• plungeai-in-<tool> — per-editor connect steps (Cursor, VS Code, Claude Code, Codex, …).
• plungeai-platform — the full capability map (agents, tools, workflows, models, missions, scheduling, memory, skills/plugins, observability) once you're connected.