CMP Wiki:For agents

This page is for software that reads cmp.wiki: crawlers, AI assistants, scripts and anyone importing the data. Everything here is public and needs no account. The text is licensed CC BY-SA 4.0: quote it, build on it, and credit the page URL and the Discord handle a fact carries.

Read a page

• Rendered page: https://cmp.wiki/wiki/Page_title. Plain HTML, no JavaScript needed, no login, no bot challenge.
• Plain text of a whole page: https://cmp.wiki/w/api.php?action=query&prop=extracts&explaintext=1&exsectionformat=plain&format=json&formatversion=2&titles=Driver_unlock
• Wikitext, with templates, credits and status badges intact: https://cmp.wiki/wiki/Driver_unlock?action=raw
• Body HTML only: https://cmp.wiki/w/api.php?action=parse&page=Driver_unlock&prop=text&format=json
• Long pages are long: Driver unlock is about 44,000 characters of text. Read one section with action=parse&section=N, or start from llms-full.txt, which has every page's lead section in one file.

Find pages

• Index, one line per page: https://cmp.wiki/llms.txt . Every page's lead in one file: https://cmp.wiki/llms-full.txt
• Search: https://cmp.wiki/w/api.php?action=query&list=search&srwhat=text&srsearch=undervolt&format=json (keep srwhat=text: without it only titles are matched)
• Sitemaps: https://cmp.wiki/sitemap-wiki/sitemap-index-cmpwiki.xml (wiki), https://cmp.wiki/wp-sitemap.xml (front door and news), https://cmp.wiki/news-sitemap.xml (news, last two days)
• What changed: https://cmp.wiki/w/api.php?action=feedrecentchanges&feedformat=atom&namespace=0 (Atom). News: https://cmp.wiki/feed/ (RSS).

Structured data

Recipes, platforms, confirmations, unlocks and seller reports are Cargo tables. Three ways in:

• Daily JSON snapshots, one file per table with the field names and the generation time: https://cmp.wiki/data/ (machine index at https://cmp.wiki/data/index.json).
• Live queries: https://cmp.wiki/w/api.php?action=cargoquery&tables=Recipe&fields=_pageName%3Dpage%2Cmodel%2Ccards%2Cdecode_single%2Cstatus&limit=500&format=json. Field lists: https://cmp.wiki/w/api.php?action=cargofields&table=Recipe&format=json.
• Export: https://cmp.wiki/wiki/Special:CargoExport?tables=Recipe&fields=_pageName%3Dpage%2Cmodel%2Ccards&format=json&limit=500 (also format=csv).

Tables and row counts on November 3, 2026: Recipe 43, Platform 23, Confirmation 59, Unlocks 11, SellerReport 110. On November 9, 2026 (the /data/index.json snapshot of 14:41 UTC): Recipe 46, Platform 24, Confirmation 63, Unlocks 11, SellerReport 110; the live counts are in /data/index.json.

The status words are explained on Status taxonomy. The computed Proven on list pages comes from Confirmation rows; the status field inside a table row is the author's own claim.

MCP server

cmp.wiki runs a read-only Model Context Protocol server for AI agents at https://cmp.wiki/mcp (Streamable HTTP, no login, no cookies). It reads the same public pages, tables and tools listed on this page, and every result names its source URL and carries the CC BY-SA 4.0 line.

• Claude Code: claude mcp add --transport http cmpwiki https://cmp.wiki/mcp
• Other clients (JSON config): {"mcpServers": {"cmpwiki": {"type": "http", "url": "https://cmp.wiki/mcp"}}}
• Opening https://cmp.wiki/mcp in a browser shows the same instructions and the tool list.

Tools:

• search: full-text search over the wiki and the news posts, with URLs and snippets.
• get_page: one page as markdown (status labels and Source credits kept) or as wikitext; long pages one section at a time.
• recipes, platforms, unlocks, confirmations and seller_reports: the structured tables from the daily snapshots, filtered. Recipe and platform status is computed from the confirmations the way the list pages do it.
• benchmarks: one row per recipe plus every posted measurement from Benchmarks, each with its source credit.
• news and recent_changes: what is new on the front door and in the wiki.
• board_open_tasks: open tasks on the public board.
• will_it_fit and card_check: the Will it fit and Card check calculations, run on the arguments or the pasted output. Nothing is stored.
• ask: Ask the wiki, an answer from the wiki's own passages with numbered sources.
• site_map: the llms.txt index in one call.

Limits: 60 requests a minute per address; ask 3 a minute and 10 an hour. Results are capped at 200,000 characters (a cut result says so and points to sections), and reads from the site time out after 60 seconds. The server never writes anything.

Manners

• Reads are welcome. Pages are cached for ten minutes at the edge, so there is no need to hammer them. Writes (edits, uploads) need a Discord sign-in and are rate limited.
• Send a User-Agent that names your project and a way to reach you.
• Summaries with a link are the norm here; do not republish whole pages without their credits. Questions, corrections or removal requests: hello@cmp.wiki or the cmpunlocker Discord.