# Anclum — AI tools, MCP servers and agent skills, with the evidence behind each entry

> Anclum (anclum.com) is an English/Chinese directory of AI tools and MCP servers. It holds 9,442 entries (5,838 MCP servers, 1,730 agent skills, 1,874 other repositories) plus 1,619 AI tools. For every entry it publishes only what it can check: how many independent sources carry it, the first and last date it was seen, whether it is remote or local together with the evidence, and whether the official link resolves. No ranking is sold: the order is stars and recency, and the page says so.

## Where to start

- [MCP server directory](https://anclum.com/en/mcp-directory/): 5,838 servers in 13 field categories, with a remote-vs-local filter and three sort orders.
- [Agent skills](https://anclum.com/en/t/skill/): 1,730 skills, grouped by field.
- [AI tools](https://anclum.com/en/ai/): 1,619 tools, filterable by free tier, Chinese UI, ICP filing, and source count >= 2.
- [By task](https://anclum.com/en/task/): 32 tasks; each page is a working guide (workflow stages, tools grouped by route, prompt templates, a checklist), not a copy of the tool list.
- [Clients & CLI](https://anclum.com/en/clients/): setup for Claude Desktop, Claude Code, Cursor, VS Code and Codex.
- [Rankings](https://anclum.com/en/best/): boards by category.
- [Link cleanliness comparison](https://anclum.com/en/link-cleanliness/): a per-cell test of whether other directories' outbound links land on the vendor's own site.

## Task guides (each page is a way to get the job done)


## AI tool categories

- [ai-office](https://anclum.com/en/ai/ai-office/)
- [ai-video](https://anclum.com/en/ai/ai-video/)
- [ai-audio](https://anclum.com/en/ai/ai-audio/)
- [ai-writing](https://anclum.com/en/ai/ai-writing/)
- [ai-agent](https://anclum.com/en/ai/ai-agent/)
- [ai-image](https://anclum.com/en/ai/ai-image/)
- [ai-design](https://anclum.com/en/ai/ai-design/)
- [ai-coding](https://anclum.com/en/ai/ai-coding/)
- [ai-platform](https://anclum.com/en/ai/ai-platform/)
- [ai-chat](https://anclum.com/en/ai/ai-chat/)
- [ai-search](https://anclum.com/en/ai/ai-search/)
- [ai-model](https://anclum.com/en/ai/ai-model/)
- [ai-detect](https://anclum.com/en/ai/ai-detect/)
- [ai-learn](https://anclum.com/en/ai/ai-learn/)
- [ai-prompt](https://anclum.com/en/ai/ai-prompt/)

## How we decide (so a model can judge whether a field is citable)

- remote vs local: only three kinds of evidence count — (1) the repository registers a remote endpoint in the official MCP Registry; (2) the entry URL is itself a streamable-http / sse endpoint; (3) the install command carries an http(s) transport. Anything else is written as "unknown" rather than guessed.
- links: followed to the final destination, tracking parameters stripped, and only the vendor's own site kept; a directory that proxies every tool through its own domain is recorded as such.
- price / free tier: stated only when the vendor's own page states it; otherwise "not stated".
- Chinese UI and ICP licence: the licence number is printed verbatim when it exists, and nothing is inferred when it does not.
- no ranking is sold; the sort rules are printed on the page.
- data refreshed daily; every entry carries a first-seen and last-seen date.

## Machine-readable entry points

- [llms-full.txt](https://anclum.com/llms-full.txt): every entry on one line (name — one-liner — page), ready to be used as retrieval corpus.
- [MCP endpoint](https://anclum.com/mcp): the same index over the MCP protocol.
- [sitemap index](https://anclum.com/sitemap.xml): the site map index.

## Counts (computed from the database at build time)

Counts: 9,442 entries = 5,838 MCP servers + 1,730 agent skills + 1,874 repositories; 1,619 AI tools; 807 remote endpoints; 32 task guides.