docs: add examples/README.md, broaden the root README's examples intro (#157)

examples/README.md was missing entirely. Add it as an index explaining
the shared theme across the three examples: each replicates a
well-known project on top of MFS, in a fraction of the code, because
MFS already owns ingestion/embedding/vector-store/search.

The root README's examples section only talked about Open Tag (a
leftover from before deep-research-skill and claude-context existed).
Reworded it to state that shared theme up front and list all three
examples evenly instead of spotlighting one.

Signed-off-by: Cheney Zhang <chen.zhang@zilliz.com>

Author: zc277584121

README.md

@@ -739,23 +739,33 @@ at MFS and focus on the app on top.
 - **CLI** — call the `mfs` command to build your own **skills**, **MCP servers**,
   or **plugins** on top of the same surface.
 
-One example is [`Open Tag`](examples/open-tag-skill/): a small open-source Slack
-bot you `@mention` — `@OpenClaude` or `@OpenCodex` — that gathers authorized
-context and runs the task on a CLI agent backend. It's an open, self-hosted homage
-to [Claude Tag](https://www.anthropic.com/news/introducing-claude-tag), Anthropic's
-hosted `@Claude` Slack teammate. The point is how little it takes to build: MFS already gives it one
-searchable, self-hosted memory over Slack, repos, docs, issues, and databases, so
-its strongest capability — plugging in whatever data sources you need — is also
-the easy part. See the [example](examples/open-tag-skill/) for the full walkthrough.
+### 🧩 Examples: well-known projects, rebuilt on MFS
+
+Every example under [`examples/`](examples/) replicates a well-known project on
+top of MFS — the same capability, in a fraction of the code — because MFS
+already owns the part that used to be the hard part for each of them:
+ingestion, chunking, embedding, the vector store, and hybrid search across many
+source types. What's left is thin glue, or a single skill describing a strategy.
+
+- [`claude-context`](examples/claude-context/) — a replica of
+  [claude-context](https://github.com/zilliztech/claude-context) ("make the
+  codebase the context for any coding agent"): an MCP server exposing
+  `search` + `read` over MFS, about 60 lines.
+- [`deep-research-skill`](examples/deep-research-skill/) — the "reason and
+  search over private data" job [deep-searcher](https://github.com/zilliztech/deep-searcher)
+  does as a standalone framework, done here as a single skill on top of
+  `mfs-find` — no vector-DB glue, no provider matrix, no custom
+  retrieval-loop code.
+- [`open-tag-skill`](examples/open-tag-skill/) — a self-hosted homage to
+  [Claude Tag](https://www.anthropic.com/news/introducing-claude-tag),
+  Anthropic's hosted `@Claude` Slack teammate: `@mention` a bot in Slack —
+  `@OpenClaude` or `@OpenCodex` — and it gathers authorized MFS context and
+  runs the task on a CLI agent backend.
 
 ![Open Tag — delegating a PR review across Slack channels](https://github.com/user-attachments/assets/6cb1db05-dd12-4a13-a9fa-1a1bf69bcf28)
 
-Another is [`Deep Research`](examples/deep-research-skill/): the "reason and
-search over private data" job [deep-searcher](https://github.com/zilliztech/deep-searcher)
-does as a standalone framework, done here as a single skill on top of `mfs-find` —
-no vector-DB glue, no provider matrix, no custom retrieval-loop code, because MFS
-already indexes and searches, and an agent with a search tool already runs that
-loop natively once it has the strategy.
+See [`examples/README.md`](examples/README.md) for the full list, or each
+example's own README for its install command and walkthrough.
 
 ## 🗺️ Roadmap
 

examples/README.md

@@ -0,0 +1,23 @@
+# Examples
+
+Each example here rebuilds a well-known project on top of MFS — the same
+capability, in a fraction of the code — because MFS already owns the part
+that used to be the hard part for each of them: ingestion, chunking,
+embedding, the vector store, and hybrid search across many source types.
+What's left is thin glue, or a single skill describing a strategy on top of
+`mfs search` / `mfs cat`.
+
+| Example | Replicates | What it is |
+|---|---|---|
+| [`claude-context/`](claude-context/) | [claude-context](https://github.com/zilliztech/claude-context) — "make the codebase the context for any coding agent" | An MCP server: `search` + `read` over MFS, about 60 lines |
+| [`deep-research-skill/`](deep-research-skill/) | [deep-searcher](https://github.com/zilliztech/deep-searcher) — reason and search over private data | A skill (`deep-research`) that runs the decompose → search → evaluate → synthesize loop on top of `mfs-find`; no framework, no vector-DB glue |
+| [`open-tag-skill/`](open-tag-skill/) | [Claude Tag](https://www.anthropic.com/news/introducing-claude-tag) — Anthropic's hosted `@Claude` Slack teammate | A self-hosted Slack bot (`@OpenClaude` / `@OpenCodex`) plus the `open-tag-admin` skill that sets it up and runs it |
+
+Every one of the replicated projects had to build its own retrieval stack
+from scratch — a provider matrix, a vector database, document loaders, sync
+logic. MFS already is that stack, so the example is only ever what's left
+on top of it: an MCP server thin enough to read in one sitting, a skill
+that's a strategy prompt over MFS's search API, or a small bridge process.
+
+Open each example's own README for the install command and a full
+walkthrough.