
Rustpad MCP Server
io.github.ni-c由 GitHub 帳號 ni-c 發布v0.4.0更新於 Nov 10, 2026
MCP server for Rustpad, the self-hosted collaborative text editor
概覽
將助理連接到自架的 Rustpad 執行個體,讓它能讀取、建立與編輯協作文字 pad。
- 可以做什麼
- 讀取 pad 及其中繼資料:純文字內容、長度、修訂版本、語言、目前編輯者與伺服器統計
- 建立 pad,可使用隨機或指定 id,並可附上內容與語法醒目提示語言
- 整份取代 pad 內容,或透過操作轉換附加與搜尋取代文字,保留他人同時進行的編輯
- 設定 pad 的 Monaco 語法醒目提示語言
- 在文字之外回傳結構化輸出,並將 pad 內容標記為不可信資料
- 適用情境
- 當你在自架的 Rustpad 執行個體中保存筆記、草稿或共用片段,並希望助理在他人可能同時編輯同一 pad 時讀取、附加或改寫內容時適用。
- 執行需求
- 需要一個可連線的 Rustpad 執行個體,透過必填的 RUSTPAD_URL 變數指定。本機執行需 Node.js 24 或更新版本,也可使用 Docker 容器。Rustpad 沒有身分驗證,因此不需帳號或憑證。選用變數可篩選工具(RUSTPAD_ALLOW_TOOLS、RUSTPAD_DENY_TOOLS)、限制為唯讀(RUSTPAD_READ_ONLY)或接受自簽憑證(RUSTPAD_INSECURE_TLS)。
安裝
在 SourceWeft 中
- 開啟 儀表板中的 Rustpad MCP Server,將其新增到工作區。
- 為需要使用其工具的對話啟用該服務。
Desktop only,透過 STDIO。 STDIO 服務會啟動本機處理程序,因此需要 SourceWeft 桌面主機。
其他 MCP 客戶端
參照 儲存庫 中的啟動說明。
README
rustpad-mcp
[CI]
[OpenSSF Scorecard]
[Socket Badge]
[Glama score]
[npm version]
[container image]
[HTTP via mcp-hub]
[docs]
[sponsor]
A Model Context Protocol (MCP) server for Rustpad, the efficient, minimal, self-hosted collaborative text editor.
Lets MCP clients like Claude Code, Claude Desktop or Codex read and write the pads of a Rustpad instance: fetch a document, create one, replace it wholesale or edit it in place.
Eight tools is the ceiling, not the floor: RUSTPAD_ALLOW_TOOLS=essential
registers a curated five instead, and a model picks the right tool far more
reliably from five than from eight — see
choosing which tools load.
Reads go through Rustpad's HTTP API; writes speak the operational-transformation
WebSocket protocol, so targeted edits (append_to_document, replace_in_document)
merge cleanly with what human collaborators type at the same time instead of
overwriting it. While the server edits a pad, it is visible to everyone in the pad as
a collaborator named rustpad-mcp.
The two edits that cannot be undone ask a person. Where the client supports
MCP elicitation, replacing a non-empty pad and search-replacing across more than
one match raise a real dialog that the model cannot answer on its behalf — and
the replace_in_document one says how many places are about to change. Where it
does not, they fall back to a two-call token, and say so rather than implying
somebody approved. ELICITATION=false takes that fallback deliberately; it
never removes the guard. See
Asking a person.
[Demo of rustpad-mcp over the MCP inspector]
[Architecture: an MCP client talks to rustpad-mcp over stdio; rustpad-mcp reads pads over HTTPS and writes them over the WebSocket OT protocol]What makes it different
Real OT edits, not overwrites. append_to_document and replace_in_document
retain everything they do not touch, and the Rustpad server transforms concurrent
edits — a human typing in the same pad at the same moment loses nothing. The model
shows up in the pad as a named collaborator.
Built for an unauthenticated world. Rustpad has no accounts, so every pad is untrusted by definition. Everything that comes out of one — reads, metadata, even upstream error bodies — is explicitly marked as data, never instructions, before a model sees it.
Requirements
- A reachable Rustpad instance (self-hosted; the server is stateless and needs no credentials — Rustpad has no authentication)
- Node.js >= 24, or Docker
Configuration
The same URL serves the HTTP API, the WebSocket endpoint and the share links
returned by the tools (<RUSTPAD_URL>/#<pad-id>). The RUSTPAD_* booleans must
be exactly true. The server starts and lists its tools without configuration;
every call then fails with setup instructions.
ELICITATION is the odd one out twice over: it carries no prefix, so it reaches
every MCP server in the same environment, and a value that is neither true nor
false stops the server rather than falling back — it is the only variable here
that defaults to on, and a typo would otherwise leave the dialog running while
you believed it was off. A server started with it off prints one line saying so.
Keep in mind what Rustpad is: pads are ephemeral (lost on server restart
and after 24 hours of inactivity, unless the instance is run with
SQLITE_URI) and anyone who knows a pad id can read and write it. Do not
put secrets in pads.
Choosing which tools load
RUSTPAD_ALLOW_TOOLS and RUSTPAD_DENY_TOOLS take comma-separated tool names;
a trailing * matches a whole family. essential is a curated preset of
five: get_document, get_document_info, create_document, set_document, append_to_document.
One boundary is softer than the list suggests: Rustpad has no create operation,
a pad exists under any id the moment it is written to, so denying
create_document does not stop new pads — set_document or
append_to_document on a fresh id makes one just the same. Deny the write
tools you mean, or use RUSTPAD_READ_ONLY.
An entry that matches no tool aborts startup and names it, so a typo cannot
silently hide a tool — an absent tool is not something anyone traces back to an
environment variable. A filtered tool is never registered, so it is absent from
tools/list and unknown to tools/call alike, exactly like a write tool under
RUSTPAD_READ_ONLY.
If you run several of these servers at once, mcp-hub
is the other answer — its /hub endpoint replaces every server's tools with six
meta-tools.
Installation
Claude Code
Claude Desktop
Codex
~/.codex/config.toml:
Docker
Through mcp-hub
A client that cannot spawn a local process — ChatGPT connectors, Claude on the web,
Cursor, LibreChat — reaches rustpad-mcp through mcp-hub: one
container serves many stdio MCP servers over Streamable HTTP, with an OAuth 2.1 login
behind a single password and long-lived tokens for the clients that cannot do OAuth. Its
/hub endpoint puts every server behind six meta-tools, so one connector reaches all of
them without N×tool schemas in the model's context, and it speaks both protocol revisions
— a question this server asks travels through it to the person at the far end.
Its /config/mcp.json uses Claude Code's format, so the entry is the one you already
have:
allowTools and denyTools there are the hub's own per-server filter, which is not
the same thing as *_ALLOW_TOOLS in env — the difference, and the mistake it invites,
are in the client guide.
Tools
👤 asks a person through MCP elicitation · falls back to a two-call
confirm_token where the client cannot show a dialog.
With RUSTPAD_READ_ONLY=true only the first three are registered.
Structured output
Every tool declares an outputSchema and answers with structuredContent
alongside the text block, so a client can use the result without parsing prose.
The five write tools used to answer with a sentence — "Appended 12 characters
to pad …" — and the sentence is still there, in the text block:
get_document answers {text} rather than the pad as the whole result, for the
same reason get_document_info has always been an object: a schema whose root
is a string is served to a 2025-era client rewritten as {result: …}, so the
tool would answer in two shapes depending on who asked. It is also where
empty and truncated can live — an empty answer used to be a sentence.
The two read tools that report pad content carry untrusted: true and
source: "rustpad" as fields. A pad is world-writable to anyone who knows its
id, including text this server wrote earlier, and a client that reads the
structured half would otherwise get it with no framing at all.
Not exposed, on purpose
No pad listing — Rustpad has no such API. Pads exist implicitly under every
id, so you have to know the ids you care about. get_stats reports how many
documents the server currently holds, but not their names.
No accounts, no permissions. Rustpad has neither, which is why every pad is treated as untrusted input rather than as something a login vouched for.
Safety
- Pad content is world-writable and therefore untrusted: every read result is prefixed with a marker telling the model to treat it as data, never as instructions.
- The two irreversible edits ask a person: a real dialog the model cannot answer on its behalf, bound to the pad and the exact replacement. Where the client cannot show one, a single-use token that only ever appears in a previous tool result — which proves the call was made twice with the same arguments, and nothing more. The fallback text says which of the two it was.
- Tool results are size-capped; control characters are stripped from everything the instance wrote, and upstream error bodies are sanitized before they reach the model.
- The instance is untrusted too: WebSocket frames are limited at the header, before they are buffered, the message queue is bounded in bytes, and every frame is checked for shape before it is folded into the pad.
RUSTPAD_INSECURE_TLSrelaxes certificate validation only for the configured connection, never process-wide.
Documentation
The full guide, tool reference and security notes live at
rustpad-mcp.ni-c.de (source in docs/).
Development
The test suite talks to an in-memory fake of rustpad-server (including OT
transformation of concurrent edits) over the real MCP protocol; no live
instance is needed. The architecture diagram and social card are generated —
edit docs/assets/architecture.source.svg and run npm run assets, never the
rendered copies.
Releasing
Releases are tag-driven. Bump package.json, move the [Unreleased] notes in
CHANGELOG.md under the new version, commit, then:
The release workflow publishes to npm via Trusted Publishing (OIDC, with provenance), pushes the multi-arch container image to GHCR, creates the GitHub release from the CHANGELOG section, and updates the entry in the official MCP registry.
Contributing
Issues, discussions and pull requests are welcome — see CONTRIBUTING.md. For vulnerabilities please use private reporting rather than a public issue; the policy is in SECURITY.md.
License
MIT © Willi Thiel
來源:README.md,提交 aee9f65
工具
0版本歷史
1- v0.4.0最新Sep 16, 2026
評分
已滿足適用檢查項的 50%
該評分衡量此收錄條目的完整程度,依據註冊中心條目、端點與公開來源計算得出。它不是對伺服器本身的評價,也不表示伺服器安全或目前可用:可連通性是帶日期的檢查結果,評分也不包含安裝量或使用者評價。
- 近 30 天內可連通25 分
請保持端點能夠回應 MCP 握手;SourceWeft 每 30 天探測一次。
- 已宣告認證方式10 分
請在註冊中心條目中宣告憑證請求標頭,或保持端點能夠回應探測,以便判定其認證方式。
- 已列出工具15 分
請讓伺服器在無憑證時回應 tools/list,或發布可在沙箱中無憑證啟動的軟體包。
- 描述必需15 分
- 圖示10 分
- 文件15 分
- 授權條款10 分