mcp-server-silverbullet
An MCP server that connects a SilverBullet space over its HTTP filesystem API, exposing a scoped prefix of notes to AI agents with read/write/delete capabilities.
README
mcp-server-silverbullet
English · 한국어
An MCP server that connects a SilverBullet space over
its HTTP filesystem API (/.fs) and exposes only a chosen prefix to AI
agents. It provides five tools plus resources, so you can attach your notes to an
agent without handing over the whole space. Runs with npx — no install needed.
Features
- Prefix scoping — the agent sees only what lives under one prefix (the folder's contents plus its namesake page). Enforced in code, not in the prompt.
- Scope gating — allow
readonly, or addwrite/delete. Tools that aren't allowed are never registered in the first place. - Path hardening — absolute paths,
.., encoded traversal, and anything outside the prefix are all rejected. - Token isolation — the SilverBullet token stays in the server environment and is never exposed to the LLM.
Note — this is a guardrail for the LLM, not a multi-tenant security boundary. SilverBullet's
SB_AUTH_TOKENgrants full space access (there are no prefix-scoped tokens), so use it only in an environment you control.
Setup
Requires Node.js ≥ 20 and a SilverBullet instance with SB_AUTH_TOKEN configured.
Add the following to your MCP client (Claude Code, Cursor, Claude Desktop, etc.):
{
"mcpServers": {
"silverbullet": {
"command": "npx",
"args": ["-y", "@bong-u/mcp-server-silverbullet@0.1.0"],
"env": {
"SILVERBULLET_MCP_BASE_URL": "https://notes.example.com",
"SILVERBULLET_MCP_TOKEN": "<SB_AUTH_TOKEN>",
"SILVERBULLET_MCP_PREFIX": "projects/work",
"SILVERBULLET_MCP_SCOPES": "read"
}
}
}
}
| Variable | Required | Default | Description |
|---|---|---|---|
SILVERBULLET_MCP_BASE_URL |
✅ | — | SilverBullet instance URL. |
SILVERBULLET_MCP_TOKEN |
✅ | — | SB_AUTH_TOKEN. Never passed to the LLM. |
SILVERBULLET_MCP_PREFIX |
— | whole space | Subpath the agent may access, e.g. projects/work. |
SILVERBULLET_MCP_SCOPES |
— | read,write,delete |
Allowed operations (must include read). |
SILVERBULLET_MCP_TIMEOUT_MS |
— | 15000 |
Request timeout (ms). |
SILVERBULLET_MCP_MAX_SEARCH_BYTES |
— | 1000000 |
Max file size search_files will scan. |
SILVERBULLET_MCP_USER_AGENT |
— | mcp-server-silverbullet/<version> |
User-Agent sent upstream. |
If a required variable is missing or malformed, the server exits immediately with a clear error.
Tools
Paths keep the prefix's top segment (notes.md, notes/todo.md). Feed a path
returned by list_files straight into read_file.
| Tool | Scope | Description |
|---|---|---|
list_files |
read |
List files under the prefix (optionally within a subpath). |
read_file |
read |
Read a file — text as utf8, binary as base64. |
write_file |
write |
Create or replace a file. |
delete_file |
delete |
Delete a file. |
search_files |
read |
Substring search over file names and text content. |
Files are also exposed as resources: silverbullet://notes/todo.md.
Development
npm install
cp .env.example .env # fill in instance URL + token
npm run dev # run straight from source (tsx), loading .env
npm run lint # biome
npm run typecheck # tsc --noEmit
npm test # vitest
npm run build # compile to dist/
CI runs lint · typecheck · test · build on Node 20, 22, and 24.
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。