cosmosdb-mcp
MCP server for Azure Cosmos DB enabling listing accounts, databases, containers, running SQL queries, and reading/writing/deleting documents.
README
Azure Cosmos DB MCP Server
An MCP (Model Context Protocol) server for Azure Cosmos DB. Compatible with any MCP client — Claude Code, Claude Desktop, Cursor, and others.
Exposes tools for listing accounts, databases, and containers, running SQL queries, reading and writing documents, and deleting items. The built-in Azure MCP server that ships with Claude Code provides limited Cosmos DB support — this project covers the full data plane, including writes.
Authentication uses DefaultAzureCredential, which picks up an active az login session automatically. Alternatively, a key can be provided via the AZURE_COSMOS_KEY environment variable. No secrets or keys are ever passed as tool arguments.
Requirements
Installation
Install uv and the Azure CLI if you don't have them:
macOS
brew install uv azure-cli
Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash # Debian/Ubuntu
For other Linux distributions see the Azure CLI install docs.
Windows
winget install --id=astral-sh.uv
winget install --id=Microsoft.AzureCLI
Configuration
Claude Code users:
claude mcp add --scope user cosmosdb -- uvx cosmosdb-mcp
For other MCP clients, add the following to your server configuration:
{
"mcpServers": {
"cosmosdb": {
"command": "uvx",
"args": ["cosmosdb-mcp"]
}
}
}
Restart your MCP client after adding the server. No environment variables are required if you are authenticated with az login. Optional env vars:
AZURE_SUBSCRIPTION_ID— used bycosmosdb_list_accountsif setAZURE_COSMOS_KEY— use instead ofaz loginfor data plane operations
Installing from source
git clone https://github.com/BrianDeacon/cosmosdb-mcp
cd cosmosdb-mcp
uv sync
az login
Then configure with the cloned path:
{
"mcpServers": {
"cosmosdb": {
"command": "uv",
"args": ["run", "--directory", "/path/to/cosmosdb-mcp", "cosmosdb-mcp"]
}
}
}
Tools
The account parameter accepts either a short account name (e.g. my-cosmos-account) or a full endpoint URL (e.g. https://my-cosmos-account.documents.azure.com). The prefix and suffix are added automatically if absent.
cosmosdb_list_accounts
List all Cosmos DB accounts in the current Azure subscription. The subscription is resolved automatically — first from the AZURE_SUBSCRIPTION_ID environment variable, then from the active az login session.
cosmosdb_list_databases
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
Returns a sorted JSON array of database names.
cosmosdb_list_containers
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
Returns a sorted JSON array of container names.
cosmosdb_get_container_info
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
container |
string | yes | Container name |
Returns partition key path, indexing policy, default TTL, unique key policy, and system properties.
cosmosdb_query_items
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
container |
string | yes | Container name |
query |
string | yes | SQL query (e.g. SELECT * FROM c WHERE c.status = 'active') |
max_items |
integer | no | Max items to return (default 100, cap 1000) |
Cross-partition queries are enabled automatically. Use cosmosdb_query_items_to_file instead if the result set may be large.
cosmosdb_query_items_to_file
Same as cosmosdb_query_items but writes results to a file. Only the item count is returned in context, avoiding large payloads filling the context window.
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
container |
string | yes | Container name |
query |
string | yes | SQL query |
output_file |
string | yes | Path to write results as a JSON array |
max_items |
integer | no | Max items to return (default 100, cap 1000) |
cosmosdb_count_items
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
container |
string | yes | Container name |
where |
string | no | SQL WHERE clause body (e.g. c.status = 'active'). If omitted, counts all items. |
Returns a JSON object with a count field.
cosmosdb_read_item
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
container |
string | yes | Container name |
item_id |
string | yes | Item id field value |
partition_key |
string | yes | Partition key value |
Returns the full item document as JSON.
cosmosdb_upsert_item
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
container |
string | yes | Container name |
item |
object | yes | Full item document — must include an id field |
Inserts or replaces the item. Returns the stored document including system fields.
cosmosdb_delete_item
Destructive. Deletes an item permanently.
| Parameter | Type | Required | Description |
|---|---|---|---|
account |
string | yes | Cosmos DB account name or endpoint |
database |
string | yes | Database name |
container |
string | yes | Container name |
item_id |
string | yes | Item id field value |
partition_key |
string | yes | Partition key value |
Security
- Authentication relies on
DefaultAzureCredentialorAZURE_COSMOS_KEY— keys are never passed as tool arguments, so secrets do not appear in conversation history. - The
accountparameter only accepts account names and endpoint URLs, never connection strings.
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。