ivedaai-mcp-server
An MCP server for the IvedaAI video analytics API, enabling natural language control of footage search, camera and alert management, analysis jobs, and watchlists.
README
ivedaai-mcp-server
An MCP server for the IvedaAI video analytics API. Point an MCP client — Claude Desktop, Claude Code, or anything else that speaks the protocol — at your IvedaAI deployment and drive it in natural language: search footage, manage cameras and alert rules, run analysis jobs, work with face and licence-plate watchlists.
316 API operations are exposed as 63 tools, one per resource type. See why.
Quickstart
Add this to your MCP client's configuration. Nothing to install first — npx fetches it.
{
"mcpServers": {
"ivedaai": {
"command": "npx",
"args": ["-y", "ivedaai-mcp-server"],
"env": {
"IVEDAAI_BASE_URL": "https://ivedaai.example.com",
"IVEDAAI_USERNAME": "your-username",
"IVEDAAI_PASSWORD": "your-password"
}
}
}
}
Where that file lives:
| client | path |
|---|---|
| Claude Desktop (macOS) | ~/Library/Application Support/Claude/claude_desktop_config.json |
| Claude Desktop (Windows) | %APPDATA%\Claude\claude_desktop_config.json |
| Claude Code | claude mcp add ivedaai --env IVEDAAI_BASE_URL=… --env IVEDAAI_USERNAME=… --env IVEDAAI_PASSWORD=… -- npx -y ivedaai-mcp-server |
Restart the client, and ask it something like "list the cameras that are currently offline".
Try it without a client:
IVEDAAI_BASE_URL=https://ivedaai.example.com \
IVEDAAI_USERNAME=you IVEDAAI_PASSWORD=secret \
npx -y ivedaai-mcp-server
It speaks JSON-RPC over stdin/stdout and logs a startup line to stderr. --help prints the
configuration reference; --version prints the version.
Read-only first
If you are evaluating this, or connecting it to anything you would not want to write to, start here:
"env": { "IVEDAAI_READ_ONLY": "true", "IVEDAAI_BASE_URL": "…", "IVEDAAI_USERNAME": "…", "IVEDAAI_PASSWORD": "…" }
Every non-GET is withheld from the tool list entirely rather than merely refused, and it roughly halves what the tool descriptions cost on connect.
Configuration
Only the first three are required.
| Variable | Default | Description |
|---|---|---|
IVEDAAI_BASE_URL |
— | Origin of your IvedaAI server, e.g. https://ivedaai.example.com. No path. |
IVEDAAI_USERNAME |
— | IvedaAI account username. |
IVEDAAI_PASSWORD |
— | IvedaAI account password. |
IVEDAAI_READ_ONLY |
false |
true serves reads only: non-GET operations are withheld from every tool, and the two write-oriented convenience tools are not registered. |
IVEDAAI_ALLOW_COLLECTION_DELETE |
false |
true permits the 20 DELETEs that name no record — see Destructive operations. |
IVEDAAI_REDACT_SECRETS |
true |
Masks credential-shaped fields (keys, secrets, passphrases) in responses. false disables it. |
IVEDAAI_ALLOW_INSECURE_TLS |
false |
true skips TLS certificate verification, for on-prem deployments with self-signed certificates. Traffic stays encrypted; the certificate is not checked. Scoped to this server's requests, not process-wide. |
IVEDAAI_TIMEOUT_MS |
30000 |
Per-request timeout, including reading the response body. Several IvedaAI endpoints block rather than failing fast when a camera is unreachable, so this matters. |
IVEDAAI_MAX_RESPONSE_BYTES |
2000000 |
Response body bytes read before truncating. |
IVEDAAI_CLIENT_ID / IVEDAAI_CLIENT_SECRET |
— | Sent as HTTP Basic auth on the token request, if your deployment requires client credentials. |
IVEDAAI_ALLOW_LOSSY_UPDATE |
false |
true disables the lossy-update guard. Intended for the maintainers' CRUD probe; leave it unset. |
IVEDAAI_SWAGGER_PATH |
bundled | Path to an alternate OpenAPI 3 document, if your deployment's API differs from the bundled one. |
Copy .env.example if you prefer a file.
Authentication
OAuth2 password grant against POST {base}/ainvr/api/oauth2/token. The server logs in on first use,
caches the access token, and refreshes it as it nears expiry. Note that the token endpoint is rate
limited: a client that starts a fresh process per request will hit it.
Destructive operations
Twenty of this API's DELETEs take no id in the path — DELETE /api/cameras versus
DELETE /api/cameras/{cameraId}. The only subject would come from an optional request body, and what
the API does when that body is omitted is not specified anywhere. One character of difference, and
the mistake cannot be undone.
They are withheld by default: absent from the tool descriptions and the operation enum, and
refused with an explanation naming the single-record alternative if a client sends one anyway. Set
IVEDAAI_ALLOW_COLLECTION_DELETE=true to permit them. IVEDAAI_READ_ONLY=true overrides that.
See SECURITY.md for the rest of the defaults, and for what leaves your deployment.
Using it
Every tool takes an operation and the arguments that operation needs:
{ "operation": "GET /api/cameras", "query": { "size": 20, "nameContains": "lobby" } }
- Usage guide — calling conventions and worked workflows: onboarding cameras, alert rules, analysis jobs, face and licence-plate watchlists.
- Tool reference — every tool, operation and parameter.
- Design and behaviour — why one tool per resource type, what the server does about partial updates the API silently discards, and the response format.
If a model needs the exact shape of a request body, ivedaai_get_schema returns it on demand rather
than every tool description carrying it.
Requirements
Node 20.18.1 or newer, and an IvedaAI 10.0 deployment. The bundled API document is 10.0; point
IVEDAAI_SWAGGER_PATH at your own if you run something else.
Contributing
See CONTRIBUTING.md. Upgrading the API document?
npm run diff:spec -- --against <new-spec.json> reports what changed and, more usefully, whether
anything this repo records now points at an operation that no longer exists.
License
MIT — see LICENSE.
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。