@rak-ad/mcp

RAK OS — media without an owner, in every country

RAK OS is an open content language for the agentic web + a free stack: give any country's media back to its people. The @rak-ad/mcp Model Context Protocol client connects any AI agent to the RAK network. Poland (rak.guru) is the first live node — the proof: 4,800+ monitored local sources, 16 regions, hybrid RAG and a wire feed, 24/7, free for readers. Your country is next — fork a node.

🕊️ Manifesto: MANIFESTO.md (PL) · 👑 The Realm: REALM.md · 🏛️ Governance: GOVERNANCE.md

🌍 Site/docs: https://rak.guru/mcp · 📦 npm: @rak-ad/mcp · 🔌 Remote: https://rak.guru/api/mcp/rak/mcp

📜 RAK language spec (v0.2): SPEC.md — the open standard "write once → own it → cited by every agent".

💸 Economics (the moat): ECONOMICS.md — pay-per-AI-citation. Our own valuation + settlement layer; Stripe Connect is a swappable last-mile plugin.

@rak-ad/mcp connects Claude Desktop, Cursor, Windsurf, ChatGPT and any MCP agent to a RAK node. The reference node today is Poland (rak.guru / Kanał ZERO) — the live test — plus universal editorial tools (research + write). The same rak_<module>_<operation> tools work for any node you run.

What is RAK OS? (TL;DR)

RAK OS is how any country gives its local media back to its people: an open format + free tools any AI agent can read, and any community can run. Poland is the first live node — the test. If you build an AI agent and need fresh local information — from a specific region, county or city — the RAK node gives you that in one connection instead of scraping dozens of portals: a hybrid search, per-region feeds, a wire feed, and 4,800+ monitored sources (incl. a curated census of 1,709 local outlets). Reader tools are free and work anonymously (no key). Your country is next — fork a node.

What you get

• 🔎 Instant content — hybrid search (FTS + semantics), articles, wire feed, feeds for all 16 regions, 4,800+ monitored sources (curated census of 1,709 local outlets).
• 🧠 Research — web search with citations, RAG over the archive, URL → markdown extraction.
• ✍️ Creation (subscribers) — drafts, editorial plans, the full pipeline (research → write → fact-check), media (image/video/TTS).
• ✅ QA (subscribers) — fact-check, moderation, uniqueness.
• 🌍 Free discovery — rak_meta_* and rak_content_* anonymously, no key. A free index of the node — Poland today, your country next.

Quickstart

Claude Desktop / Cursor / Windsurf — mcp.json

{
  "mcpServers": {
    "rak": {
      "command": "npx",
      "args": ["-y", "@rak-ad/mcp"],
      "env": {
        "RAK_API_KEY": "rk_...",        // omit for anonymous (reader) access
        "RAK_BASE_URL": "https://rak.guru"
      }
    }
  }
}

Remote (server agents) — Streamable HTTP

POST https://rak.guru/api/mcp/rak/mcp
Authorization: Bearer <RAK_API_KEY>     # omit for the anon tier
Api-Version: 2026-05

Inspect / smoke

npx @modelcontextprotocol/inspector

Access tiers

Tier	Key	Scope
anon (reader)	no	content_*, rag_*, voice_*, legal_*, meta_* + web research / URL extract (daily caps)
free	yes	as above + higher limits
paid (subscriber)	yes	+ writing, media, fact-check, deep research (on credits)
partner	yes	read-only (content + RAG), no research
internal	yes	full (publishing + crawl + distribution)

Free vs paid — exactly what costs what

Reading is free, forever, and needs no key. Only making things costs credits.

Free, no key (anon — 16 of the 44 tools): content_search, content_get_article, content_list_section, content_wire_feed, content_region_feed, rag_find_related, rag_semantic_search, rag_search_chunks, voice_search, legal_search, meta_list_sources, meta_list_skills, meta_health, meta_list_agents, plus research_web (5/day) and research_extract (15/day). Zero credits. 30 requests/min per tool, per IP. Cache hits never count.

Free but needs a key: owned_publish / owned_list / owned_get / owned_verify (your own content + provenance), qa_moderate, qa_uniqueness, write_plan, write_export — 0 credits on a paid tier key.

Costs credits (the creative layer, paid key):

Tool	Credits
content_create_poll	0.05
research_fact_pack	0.10
qa_fact_check	0.15
research_extract · research_web (subscribers only)	0.20 · 0.70
research_summarize	0.70
research_deep	1.10
write_edit	0.40
write_draft · write_debate · write_document	0.80
write_pipeline (per run)	1.50 news · 3.00 deep dive
media_tts	0.25
media_generate_image	0.55 (after the monthly quota)
media_storyboard	1.50
media_generate_song	4.00
media_generate_video	5.50–7.50 (after the monthly quota)

Internal only (node operators): write_publish, crawl_*, distribution_publish.

Getting a key: self-service after your first credit top-up (max 5 active keys) — rk_ keys are tier paid, scopes content:read + skills:action. Reader tools never need one.

Honest note on privacy: every tool call is logged for the attribution ledger (this is what makes pay-per-AI-citation possible) and for audit. Nothing about readers of the portal — this is the agent API.

Tool catalog (rak_<module>_<op>) — 44 tools

Authoritative surface: rak_meta_list_skills / schemas/tools.manifest.json.

Module	Tools	Tier
content	search, get_article, list_section, wire_feed, region_feed · create_poll (paid)	anon (reader)
rag	find_related, semantic_search, search_chunks	anon (reader)
voice	search — persona-knowledge RAG (a public figure's own words, source-backed)	anon (reader)
legal	search — Polish-law corpus (ELI/ISAP), citable statute fragments	anon (reader)
meta	list_sources, list_skills, health, list_agents	anon / discovery
research	web, extract, fact_pack, summarize, deep	anon → paid
write	draft, edit, plan, pipeline, publish, export, debate, document	paid → internal
media	generate_image, generate_video, generate_song, storyboard, tts	paid
qa	fact_check, moderate, uniqueness	paid
owned	publish, list, get, verify	creator (paid+)
crawl / distribution	crawl_search, crawl_add_source, crawl_subscribe, distribution_publish	internal

Per-skill details: skills/.

Examples (for agents)

• Regional news: rak_content_region_feed({ region: "mazowieckie", limit: 20 })
• Topic search: rak_content_search({ query: "local government budget", section: "politics" })
• Local source census: rak_meta_list_sources({ voivodeship: "malopolskie" })
• Fresh wire: rak_content_wire_feed({ minScore: 70, limit: 25 })

FAQ

Is RAK OS free? Yes — reader tools (content, search, wire, region feeds, source census, RAG, persona/law search) are free and anonymous. Only the creative RAK tools for subscribers are paid.

How fresh is the data? The pipeline harvests and processes sources 24/7, in 10–15 minute cycles. The rak://health resource shows freshness and 24h volume.

How many sources? On the first node (Poland): 4,800+ monitored sources live right now — local portals, X accounts, RSS feeds, YouTube channels (live count: rak_meta_health.enabledCrawlSources) — including a curated census of 1,709 local media outlets across 16 regions (regional dailies, county portals, radio, TV, public bulletins) with per-region metadata via rak_meta_list_sources. Every country's node runs its own census.

How is it different from RSS/an API? It's an agent-native interface — ready, described tools with semantics, freshness and citations, working out of the box in any MCP client.

Commercial use? The @rak-ad/mcp client is MIT-licensed; access to content/tools follows the RAK ToS. Dedicated keys for partners/enterprise.

API keys

Reader tools need no key. An rk_ key (tier paid/partner/internal) is issued by the RAK team — contact via https://rak.guru. Scopes: content:read, skills:action. Keys are tenant-scoped.

Links

• Site / docs: https://rak.guru/mcp
• LLM context: https://rak.guru/llms.txt
• Portal: https://rak.guru

Who's behind this

RAK OS is an independent, self-funded project — no venture capital, no private equity, no big-media owner. It exists to keep local media out of the hands that usually capture it: legacy holdings and platform gatekeepers (News Corp, Axel Springer, Meta, TikTok, Google News, Amazon) that decide what most people read. On scanning mandates like Chat Control, the stance is plain: we would leave a market before wiring a scanning backdoor into a reader — see the Manifesto. It is a kingdom, honestly: one maintainer (the Crown, @Hei33enberg) holds the vision and the last word; the community builds and is credited in the open (REALM.md). Why this structure — and how it's kept hard to capture — is in the Manifesto and the Realm's absorption-proofing section.

Build your own RAK

The OS is MIT — fork your country's RAK, for free, forever.

• Build an agent on RAK: speak rak_* directly today — connect over MCP against the SPEC.md, share one base and one citation market. Ship it and it's discoverable via rak_meta_list_agents. A one-command npx create-rak-agent scaffold is on the roadmap.
• Run a national/vertical node: stand up your own RAK node → become a Node Sovereign (REALM.md, L5). The RAK trademark and origin brand stay with the Crown (TRADEMARK.md); everything else is yours.
• Start from SPEC.md (the contract) and the skills.

Built on Open Mercato

RAK runs on Open Mercato — an open, inspectable runtime engine. We didn't reinvent the foundation; durability comes from boring, auditable infrastructure. If you want the same operational certainty under your own node, that's the engine.

Community

• Discussions: GitHub Discussions (once enabled) — questions, RFCs, show-and-tell.
• Contribute: CONTRIBUTING.md — humans and AI agents both (AGENTS.md).
• Chat: the RAK app ships a built-in community chat as an in-app client feature — part of the app, not the public MCP toolset.

License

MIT (client). Access to content/API follows the RAK ToS.

_{Tags: open content language · media without an owner · local news for AI agents · agentic web · content provenance · pay-per-AI-citation · model context protocol · RAG · fork a node · Poland (first node) · Kanał ZERO · rak.guru · Claude · Cursor · Perplexity}