mcp-server-db
A universal MCP server enabling AI assistants to query and manage six database engines (Postgres, Redis, Elasticsearch, MySQL, MongoDB, LDAP) through 113+ tools with read-only safety and fault tolerance.
README
mcp-server-db
mcp-server-db is a universal, production-grade Model Context Protocol (MCP) server that empowers AI assistants to interact with Postgres, Redis, Elasticsearch, MySQL, MongoDB, and LDAP through a unified suite of 113+ high-performance tools.
Built with resilience in mind, it features background lazy connection management, exponential backoff retries, circuit breaker patterns, strict read-only security enforcement, and on-demand dynamic driver loading.
Why mcp-server-db?
- Multi-Database Support: Connect to Postgres, Redis, Elasticsearch (v7, v8 & v9), MySQL, MongoDB, and LDAP directories simultaneously.
- On-Demand Light Drivers: Zero bloat. Drivers (
pg,ioredis,mongodb,mysql2,@elastic/elasticsearch,ldapts) are dynamically loaded only when a connection to that database engine is initialized. - Ironclad Safety: Independent per-connection read-only mode with session-level enforcement for relational databases and query result truncation guards to prevent LLM context overflows.
- Fault-Tolerant Architecture: Server starts instantly regardless of database availability. Dead or unreachable databases fail gracefully without crashing the MCP process or blocking other connections.
- Multi-Connection Routing: Query multiple environments (e.g., production read-only replica + local scratch cache) in parallel within the same session.
Supported Database Engines
| Database Engine | Default Driver | Supported Versions | Key Features & Guardrails |
|---|---|---|---|
| PostgreSQL | pg (optional) |
10+ | Table/Index stats, Health diagnostics, Session-level read_only, Statement timeouts |
| Redis | ioredis (optional) |
5+ | Key/String/Hash/Set/List operations, Automatic 1000-item result truncation cap |
| Elasticsearch | @elastic/elasticsearch / es7-client |
7.x, 8.x, 9.x | Dual-API version facade (v7 legacy & v8/v9 modern), Cluster health, Query DSL |
| MySQL / MariaDB | mysql2 (optional) |
5.7+, 8.0+ | DDL extraction (SHOW CREATE), Processlist management, Indexes & Triggers, Session read-only |
| MongoDB | mongodb (optional) |
4.4+ | Aggregation pipelines, Distinct queries, Index management, Execution plan explain |
| LDAP | ldapts (optional) |
v3 directories (OpenLDAP, AD, etc.) | Search/compare, Add/modify/delete entries, Bulk entry seeding |
Quick Start
1. Installation & Optional Drivers
When installed globally (npm install -g mcp-server-db) or run via npx -y mcp-server-db, npm automatically includes all supported database drivers (pg, ioredis, @elastic/elasticsearch, mysql2, mongodb, ldapts) out of the box via optionalDependencies.
# Global installation (includes all drivers out of the box)
npm install -g mcp-server-db
# Local project installation
npm install mcp-server-db
If you install mcp-server-db locally using --no-optional to keep your node_modules lightweight, you can install only the specific drivers you need:
# PostgreSQL
npm install pg
# Redis
npm install ioredis
# Elasticsearch
npm install @elastic/elasticsearch # For v8 & v9
npm install es7-client # For v7 legacy
# MySQL / MariaDB
npm install mysql2
# MongoDB
npm install mongodb
# LDAP
npm install ldapts
2. Zero-Install Execution via NPX
Add mcp-server-db directly to your MCP client config file (e.g. Claude Desktop, Claude Code, Gemini CLI, Cursor):
{
"mcpServers": {
"mcp-server-db": {
"type": "stdio",
"command": "npx",
"args": ["-y", "mcp-server-db"],
"env": {
"POSTGRES_URL": "postgresql://user:password@localhost:5432/mydb",
"REDIS_URL": "redis://localhost:6379",
"MONGODB_URL": "mongodb://localhost:27017",
"LDAP_URL": "ldap://localhost:389",
"LDAP_BIND_DN": "cn=admin,dc=example,dc=com",
"LDAP_BIND_PASSWORD": "admin-password"
}
}
}
}
3. Fast Setup with Environment Variables
Define connection strings in env to start immediately without a configuration file:
POSTGRES_URL=postgresql://user:password@localhost:5432/mydb
REDIS_URL=redis://localhost:6379
ELASTICSEARCH_URL=http://localhost:9200
MYSQL_URL=mysql://user:password@localhost:3306/mydb
MONGODB_URL=mongodb://localhost:27017
LDAP_URL=ldap://localhost:389
LDAP_BIND_DN=cn=admin,dc=example,dc=com
LDAP_BIND_PASSWORD=admin-password
POSTGRES_READ_ONLY=true
REDIS_READ_ONLY=true
MYSQL_READ_ONLY=true
MONGODB_READ_ONLY=true
LDAP_READ_ONLY=true
Complete Tools Reference (113 Tools)
System & Management Tools (2)
db_status: Inspect connection states (idle, connecting, connected, retrying, failed, circuit_open), read-only mode, and error logs across all configured databases.db_reload_config: Hot-reloaddatabases.config.ymlfrom disk and reconcile connections dynamically without interrupting active queries.
Postgres Tools (30)
pg_query: Execute read-only SELECT queries (enforces single-statement extended query protocol).pg_execute_sql: Execute write SQL statements (INSERT, UPDATE, DELETE, DDL). Blocked in read-only mode.pg_list_tables: List tables in a target schema.pg_describe_table: Inspect column data types, ordinal position, and nullability.pg_list_schemas: List all database schemas.pg_list_views: List all views in a schema.pg_list_indexes: List indexes and index definitions for a table.pg_list_triggers: List triggers and action statements in a schema.pg_table_stats: Get total size, table size, index size, and estimated row count.pg_list_constraints: List primary keys, foreign keys, and check constraints.pg_explain_query: Inspect query execution plans with EXPLAIN.pg_active_queries: Retrieve running non-idle queries and durations frompg_stat_activity.pg_list_functions: List user-defined functions, arguments, and return types.pg_list_sequences: List sequence objects in a schema.pg_database_info: Get database name, current user, PostgreSQL version, and database size.pg_kill_query: Terminate a running backend process by PID. Blocked in read-only mode.pg_vacuum_analyze: Run VACUUM ANALYZE to update statistics and reclaim dead tuples. Blocked in read-only mode.pg_list_materialized_views: List materialized views in a schema.pg_refresh_materialized_view: Refresh a materialized view (supports CONCURRENTLY). Blocked in read-only mode.pg_index_usage: Inspect index scan counts and tuple fetch statistics.pg_lock_info: Retrieve active database locks and query durations.pg_get_top_queries: Get slow queries frompg_stat_statements.pg_explain_hypothetical_index: Simulate hypothetical index performance usinghypopg.pg_database_health: Run health audit (connections utilization, buffer cache hit ratio, dead rows, replication lag).pg_sizes_overview: Breakdown storage sizes for database, tables, and indexes.pg_unused_indexes: Identify zero-scan candidate indexes to drop.pg_duplicate_indexes: Find overlapping or redundant indexes.pg_long_running_queries: Find queries exceeding a duration threshold.pg_missing_indexes: Identify high-sequential-scan tables needing indexes.pg_bloat_estimate: Estimate table bloat ratio based on dead tuple count.
Redis Tools (27)
redis_get: Get string value at key.redis_set: Set key value. Blocked in read-only mode.redis_mset: Set multiple key/value pairs in a single command (e.g. for seeding). Blocked in read-only mode.redis_del: Delete key. Blocked in read-only mode.redis_keys: List keys matching a glob pattern (capped at 1000 items).redis_ttl: Get remaining TTL in seconds.redis_hget: Get hash field value.redis_hset: Set hash field value. Blocked in read-only mode.redis_hdel: Delete hash fields. Blocked in read-only mode.redis_hgetall: Get all hash fields and values.redis_hexists: Check hash field existence.redis_sadd: Add member to set. Blocked in read-only mode.redis_srem: Remove member from set. Blocked in read-only mode.redis_smembers: List set members (capped at 1000 items).redis_sismember: Test set membership.redis_lpush: Prepend element to list. Blocked in read-only mode.redis_rpush: Append element to list. Blocked in read-only mode.redis_lpop: Remove and return first list element. Blocked in read-only mode.redis_rpop: Remove and return last list element. Blocked in read-only mode.redis_lrange: Get range of list elements (capped at 1000 items).redis_llen: Get list length.redis_exists: Check key existence.redis_expire: Set key expiration in seconds. Blocked in read-only mode.redis_type: Retrieve data type of key.redis_incr: Increment integer key. Blocked in read-only mode.redis_decr: Decrement integer key. Blocked in read-only mode.redis_flushdb: Clear current Redis database. Blocked in read-only mode.
Elasticsearch Tools (11)
es_cluster_health: Get status (green/yellow/red), node counts, and shard metrics.es_list_indices: List indices, doc counts, store size, and status.es_index_stats: Get detailed metrics for a specific index.es_search: Execute search using Query DSL.es_count: Count documents matching a query.es_get_doc: Fetch document by ID.es_index_doc: Index or overwrite document. Blocked in read-only mode.es_bulk_index: Index multiple documents into one index in a single_bulkrequest, with per-document success/failure reporting (e.g. for seeding). Blocked in read-only mode.es_update_doc: Partially update document fields. Blocked in read-only mode.es_delete_doc: Delete document by ID. Blocked in read-only mode.es_delete_by_query: Delete documents matching Query DSL. Blocked in read-only mode.
MySQL Tools (18)
mysql_query: Execute read-only SELECT/SHOW/EXPLAIN queries.mysql_execute_sql: Execute write SQL (INSERT, UPDATE, DELETE, DDL). Blocked in read-only mode.mysql_list_databases: List databases on server.mysql_list_tables: List tables in database.mysql_list_views: List views and security options.mysql_describe_table: Show column definitions and metadata.mysql_show_create_table: Extract rawCREATE TABLEDDL statement.mysql_show_create_view: Extract rawCREATE VIEWDDL statement.mysql_list_indexes: Show table indexes (SHOW INDEX).mysql_list_triggers: List database triggers.mysql_list_routines: List stored procedures and functions.mysql_list_constraints: List foreign keys and constraints.mysql_table_stats: Retrieve estimated rows, data length, and index size.mysql_active_queries: Retrieve running threads (SHOW FULL PROCESSLIST).mysql_kill_query: Kill running process by ID. Blocked in read-only mode.mysql_explain_query: Analyze query execution plan (EXPLAIN).mysql_global_status: Query server global status metrics.mysql_analyze_table: RunANALYZE TABLEto refresh statistics. Blocked in read-only mode.
MongoDB Tools (19)
mongo_list_databases: List databases on MongoDB cluster.mongo_list_collections: List collections in database.mongo_find: Query documents with filter, projection, sort, limit, and skip.mongo_distinct: Get unique field values.mongo_aggregate: Execute MongoDB aggregation pipeline.mongo_count_documents: Count documents matching filter.mongo_insert_one: Insert single document. Blocked in read-only mode.mongo_insert_many: Insert array of documents. Blocked in read-only mode.mongo_update_one: Update single document matching filter. Blocked in read-only mode.mongo_update_many: Update multiple documents. Blocked in read-only mode.mongo_delete_one: Delete single document. Blocked in read-only mode.mongo_delete_many: Delete multiple documents. Blocked in read-only mode.mongo_create_index: Create collection index. Blocked in read-only mode.mongo_drop_index: Drop collection index by name. Blocked in read-only mode.mongo_list_indexes: List all collection indexes.mongo_db_stats: Get database storage statistics.mongo_collection_stats: Get storage, index size, and object count for collection.mongo_explain: Get query planner execution plan and index scan metrics.mongo_server_status: Get cluster memory, connection, and operation stats.
LDAP Tools (6)
ldap_search: Search a directory by base DN, filter, and scope (capped at 500 entries).ldap_compare: Compare an attribute/value pair against an entry.ldap_add: Create a new entry. Blocked in read-only mode.ldap_add_bulk: Create multiple entries in one call (e.g. for seeding), with per-entry success/failure reporting. Blocked in read-only mode.ldap_modify: Add, replace, or delete an attribute's values on an entry. Blocked in read-only mode.ldap_delete: Delete an entry. Blocked in read-only mode.
Multi-Connection YAML Configuration
For complex setups with multiple named databases, create config/databases.config.yml (copy config/databases.config.example.yml):
connections:
- id: primary-pg
type: postgres
connectionString: ${POSTGRES_URL}
defaultSchema: public
statementTimeoutMs: 30000
readOnly: true
- id: app-cache
type: redis
connectionString: ${REDIS_URL}
readOnly: false
- id: search-logs
type: elasticsearch
connectionString: ${ELASTICSEARCH_URL}
apiVersion: "9"
readOnly: true
- id: legacy-cluster
type: elasticsearch
connectionString: ${LEGACY_ELASTICSEARCH_URL}
apiVersion: "7"
readOnly: true
- id: analytics-mysql
type: mysql
connectionString: ${MYSQL_URL}
statementTimeoutMs: 30000
readOnly: true
- id: document-store
type: mongodb
connectionString: ${MONGODB_URL}
defaultDatabase: production
readOnly: true
- id: directory
type: ldap
connectionString: ${LDAP_URL}
bindDn: ${LDAP_BIND_DN}
bindPassword: ${LDAP_BIND_PASSWORD}
readOnly: false
Local Development & Custom Builds
1. Build and Run Locally
git clone https://github.com/Nam088/mcp-database-server-v2.git
cd mcp-database-server
npm install
npm run build # Compiles TypeScript to dist/
npm test # Runs Vitest test suite
2. Configure Local Build in MCP Clients
{
"mcpServers": {
"mcp-server-db": {
"type": "stdio",
"command": "node",
"args": ["/absolute/path/to/mcp-database-server/dist/index.js"],
"env": {
"DATABASES_CONFIG_PATH": "/absolute/path/to/mcp-database-server/config/databases.config.yml"
}
}
}
}
3. Claude Code CLI
claude mcp add --scope user mcp-server-db node /absolute/path/to/mcp-database-server/dist/index.js --env DATABASES_CONFIG_PATH=/absolute/path/to/mcp-database-server/config/databases.config.yml
Client Config Paths
- Claude Desktop:
~/Library/Application Support/Claude/claude_desktop_config.json - Claude Code (Global):
~/.claude.json - Gemini CLI / IDE:
~/.gemini/config/mcp_config.json - Cursor / Project Root:
.mcp.json
License
Distributed under the MIT License. See LICENSE for details.
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。