
JSON DB MCP Server
Enables users to manage data in a simple JSON file database through MCP tools and REST API. Supports creating, reading, updating, and deleting items organized in collections with auto-generated UUIDs.
README
JSON DB MCP Server (FastAPI + FastMCP)
A minimal Model Context Protocol (MCP) server that also exposes a small REST API.
Data is stored in a single JSON file (db.json
) as collections of items with auto-generated UUIDs.
- MCP Tools:
db.add_item
,db.get_item
,db.list_items
,db.update_item
,db.delete_item
- REST Endpoints:
POST/GET/PUT/DELETE /db/{collection}[/{id}]
- Transports: HTTP (recommended) and SSE (optional)
- Stack: Python, FastAPI, Uvicorn, FastMCP
Features
- Simple JSON-file database with atomic writes
- Thread-safe in a single-process server using a lock
- Shared core logic for both MCP tools and REST
- Works with VS Code Copilot (Chat) and Gemini CLI as an MCP server
Requirements
- Python 3.10+
uv
(recommended) orpip
- Git (optional, for version control)
Project Layout
.
├─ main.py
├─ db.json # created on first write
└─ README.md
Installation
Using uv
(recommended)
uv venv
uv pip install fastapi uvicorn fastmcp
Using pip
python -m venv .venv
source .venv/bin/activate
pip install fastapi uvicorn fastmcp
Running
Development (auto-reload)
uv run main.py
# or
uv run python main.py
# Server: http://localhost:8000
Default MCP transport: HTTP at
/mcp/
(recommended).
If you prefer SSE, see MCP over SSE (optional) below.
REST API
All endpoints operate on a collection (created on first write) with items of shape:
{
"id": "uuid-string",
"...": "fields you provide"
}
Create
curl -X POST http://localhost:8000/db/users \
-H "Content-Type: application/json" \
-d '{"name":"Omar","role":"admin"}'
List
curl http://localhost:8000/db/users
Get by ID
curl http://localhost:8000/db/users/<ITEM_ID>
Update
curl -X PUT http://localhost:8000/db/users/<ITEM_ID> \
-H "Content-Type: application/json" \
-d '{"role":"owner"}'
Delete
curl -X DELETE http://localhost:8000/db/users/<ITEM_ID>
MCP Server (HTTP transport, recommended)
This server mounts FastMCP’s HTTP transport at /mcp/
.
VS Code — GitHub Copilot (Chat)
- Open Copilot Chat → Tools (or Add MCP).
- Transport:
HTTP
- URL:
http://localhost:8000/mcp/
(include the trailing slash) - Enable the server. You should see tools:
db.add_item
,db.get_item
,db.list_items
,db.update_item
,db.delete_item
Using tools in chat:
Select a tool from the list or ask:
Use
db.add_item
withcollection="users"
andpayload={"name":"Omar"}
.
Gemini CLI
Add the server via CLI (writes settings for you):
# Project-scoped (./.gemini/settings.json)
gemini mcp add --transport http jsondb http://localhost:8000/mcp/
# Or user-scoped (~/.gemini/settings.json)
gemini mcp add --scope user --transport http jsondb http://localhost:8000/mcp/
List tools:
gemini mcp list
Call a tool:
gemini mcp call jsondb db.add_item \
--args collection=users \
--args payload='{"name":"Omar"}'
MCP over SSE (optional)
If you prefer SSE, change the mount in main.py
:
# replace the HTTP mount:
# mcp_http_app = mcp.http_app(path="/")
# app.mount("/mcp", mcp_http_app)
mcp_sse_app = mcp.sse_app()
app.mount("/mcp", mcp_sse_app) # SSE lives under /mcp/sse/
- SSE endpoint:
http://localhost:8000/mcp/sse/
(note trailing slash) - Copilot: Transport =
SSE
, URL =http://localhost:8000/mcp/sse/
- Gemini CLI:
gemini mcp add --scope user --transport sse jsondb http://localhost:8000/mcp/sse/
Many HTTP clients require a trailing slash for SSE endpoints; use
/mcp/sse/
.
Health Check (optional)
Add this to main.py
(after app = FastAPI(...)
) if you want a friendly root:
@app.get("/")
def health():
return {"ok": True, "rest": "/db/{collection}", "mcp": "/mcp/"}
Persistence & Backups
- Data file:
db.json
(created on first write). - Writes are atomic: a
db.json.tmp
is written then replaced. - For safety, consider periodic snapshots (e.g.,
cp db.json db.json.bak
in a cron job).
Authentication (optional)
For local development MCP/REST are open.
If you need auth, simplest is an API key header via FastAPI middleware and headers in your MCP client configuration. Example:
from fastapi import Request
from fastapi.responses import JSONResponse
import os
API_KEY = os.getenv("API_KEY", "")
@app.middleware("http")
async def api_key_guard(request: Request, call_next):
if API_KEY and request.headers.get("x-api-key") != API_KEY:
return JSONResponse(status_code=401, content={"detail": "Unauthorized"})
return await call_next(request)
- Copilot/Gemini: configure an
x-api-key
header in the MCP server settings.
Troubleshooting
-
404 on
/
Expected. Use REST under/db/...
and MCP at/mcp/
(HTTP) or/mcp/sse/
(SSE). -
307 redirect from
/mcp
→/mcp/
Some SSE clients don’t follow redirects. Use the trailing slash or disable redirect slashes:app.router.redirect_slashes = False
-
SSE returns 404
Make sure you’re hitting/mcp/sse/
(not/mcp/
). -
Concurrency
Run as a single process for file-backed DB. Multi-process/worker deployments need a real DB or OS-level file locks. -
Git push errors (VS Code AskPass socket)
Unset helper vars or use GitHub CLI / SSH keys:unset GIT_ASKPASS SSH_ASKPASS VSCODE_GIT_ASKPASS_NODE VSCODE_GIT_ASKPASS_MAIN
License
Copyright© - Omar SOLIMAN
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。