MCP Everything (Reference Test Server) 🛠
:material-flask-outline:
examples · community ga L2
The official MCP reference test server, maintained by the MCP working group. Implements **every** protocol primitive (tools / resources / prompts / sampling / completion / logging / progress / root listing) - including the ones server authors usually skip. The catalog ships per-OS STDIO entries; the in-app QA recipe uses the Docker form as a smoke test for the playground's own MCP client.
**Vendor** `modelcontextprotocol` · T2 vendor
**Auth** None - connects without credentials
Click for full protocol-coverage breakdown · per-OS command · Docker alternative
**Vendor** - `modelcontextprotocol` (vendor-official reference; the project that defines the MCP spec itself)
**Transport** - STDIO
**Command + Args (per OS)**
| OS | Command | Arguments |
|---|---|---|
| macOS | `npx` | `-y @modelcontextprotocol/server-everything` |
| Linux | `npx` | `-y @modelcontextprotocol/server-everything` |
| Windows | `npx.cmd` | `-y @modelcontextprotocol/server-everything` |
**Auth** - None - connects without credentials.
**Prereq** - Node.js 18+ on the host so `npx` (or `npx.cmd`) can fetch and spawn the package. The catalog automatically picks the entry matching the host OS.
**Docker alternative** - if you'd rather not install Node, switch the form's **Command** to `docker` and **Arguments** to `["run", "-i", "--rm", "tzolov/mcp-everything-server:v2"]` before saving. This is the same form the playground's QA recipe uses as a smoke test.
**Why this server is catalogued under Examples**
MCP Everything is the only server in the ecosystem that *intentionally* implements every protocol primitive. That makes it the **"does my MCP client wire every primitive end-to-end?"** check for any new client - including Spring AI Playground itself. After Save & Connect:
| Inspector tab | What MCP Everything exposes |
|---|---|
| **Tools** | 15 tools, shown in the Inspector as **Echo Tool**, **Get Sum Tool**, **Print Environment Tool**, **Get Tiny Image Tool**, **GZip File as Resource Tool**, **Get Annotated Message Tool**, **Get Resource Links Tool**, **Get Resource Reference Tool**, **Get Structured Content Tool**, **Trigger Long Running Operation Tool**, **Trigger Sampling Request Tool**, **Trigger Elicitation Request Tool**, **Toggle Simulated Logging**, **Toggle Subscriber Updates**, **Simulate Research Query** - different argument schemas, progress notifications, and content shapes. As a vendor-official STDIO server they each compose to L2 - Low; confirm the live set and exact tool names on the connection's [Inspector](/spring-ai-playground/docs/features/mcp-server/inspector.html#tools) |
| **Resources** | static resources (`test://static/resource/1` ... `100`) plus parameterised resource templates (`demo://resource/dynamic/text/{resourceId}` and `demo://resource/dynamic/blob/{resourceId}`) that render against the JSON-Schema-typed input panel |
| **Prompts** | `simple-prompt` (no args), `args-prompt` (city + state), `completable-prompt` (Team Management with an MCP-completion-backed `department` argument) |
| **Notifications** | live feed of `notifications/message`, `resources/updated`, `resources/list_changed`, `prompts/list_changed`, `tools/list_changed` - toggle on with `Toggle Subscriber Updates` to start the periodic resource-updated stream |
| **Sampling** | `Trigger Sampling Request Tool` invokes `sampling/createMessage` back into the playground (client primitive); the request lands as a Sampling-tab card with Approve / Reject |
| **Elicitation** | `Trigger Elicitation Request Tool` invokes `elicitation/create` (client primitive); the Elicitation tab renders the requested form |
| **Roots** | calls `roots/list` on startup - the Roots tab shows what was asked; if no roots are advertised, the server proceeds anyway |
| **Logging** | structured log records the Inspector folds into Notifications (`Toggle Simulated Logging` toggles the cadence) |
| **Completion** | `completion/complete` for argument auto-complete inside the Inspector (`completable-prompt`'s `department` argument shows this) |
If any of these don't behave as documented, the bug is almost certainly in the playground's client implementation, not in the server - which is the whole reason MCP Everything is shipped as the test target.
**Step-by-step walkthrough** - drive every row of the table above against a live MCP Everything connection in [Tutorial 9: MCP Everything - All 8 Primitives in One Walkthrough](/spring-ai-playground/docs/tutorials/9-mcp-everything.html). Sample inputs (Echo, Add, Annotated Message), expected outputs (text + base64 image content), and the Sampling / Elicitation Approve-Reject cards are all spelled out there.
!!! tip "Hosted OAuth variant is intentionally not catalogued"
The MCP working group also publishes a Streamable HTTP variant of MCP Everything at `https://example-server.modelcontextprotocol.io/` that requires an OAuth dance. That endpoint exists for spec-conformance demos and isn't included as a catalog entry - first-time users should hit the STDIO variant which needs no credentials at all. If you want to exercise the OAuth flow specifically, add a Custom Server entry pointing at that URL with the OAuth 2.1 preset.
**Description (full)**
[macOS] Official MCP reference test server exercising every protocol feature - tools, resources, prompts, sampling, completion, logging, progress, root listing.
The activated form is pre-filled to run: `npx -y @modelcontextprotocol/server-everything`. Docker alternative: `docker run -i --rm tzolov/mcp-everything-server:v2`. The hosted `example-server.modelcontextprotocol.io` endpoint requires OAuth and is not auto-activated.
**Docs** - [github.com/modelcontextprotocol/servers/tree/main/src/everything](https://github.com/modelcontextprotocol/servers/tree/main/src/everything)
DeepWiki 🆓
:material-book-search-outline:
examples · free-tier ga L2
Cognition's hosted, free, no-auth Streamable HTTP server that answers natural-language questions about any public GitHub repo, backed by auto-generated wikis. Catalogued under Examples because it's primarily useful as a *reference grounding* server for AI code generation - a concrete external MCP target with no signup friction.
**Vendor** Cognition · T2 vendor
**Auth** None - free, no API key
Click for transport · description · docs
**Vendor** - Cognition (vendor-official; the company behind Devin)
**Transport** - Streamable HTTP
**URL** - vendor-managed; the catalog entry carries the current production URL
**Auth** - None
**Cost** - Free
**Required env** - -
**Tools** - discovered on connect; the three this server publishes (all read-only, so each scores L2 - Low on its own connection's Inspector):
??? abstract "Tools (3) - read_wiki_structure · read_wiki_contents · ask_question"
- **`read_wiki_structure`** - list the documentation topics / sections for a repo. L2 - Low
- **`read_wiki_contents`** - fetch the full generated wiki text for a repo. L2 - Low
- **`ask_question`** - ask a natural-language question about a repo. L2 - Low
**Why this server is catalogued under Examples**
DeepWiki is catalogued under Examples rather than Search because its primary use is as a *reference grounding* server for AI code generation - the kind of MCP target a tutorial points at when it wants a concrete external server with no signup friction. After Save & Connect the Tools tab exposes its three repo-wiki read tools - useful for verifying chat-side tool wiring without burning a vendor quota.
**Description (full)**
Free, no-auth Streamable HTTP server by Cognition that answers natural-language questions about any public GitHub repo, backed by auto-generated wikis. Three read-only tools - `read_wiki_structure`, `read_wiki_contents`, `ask_question`.
**Docs** - [deepwiki.com](https://deepwiki.com)