Best MCP Servers for Database Access
MCP servers that connect LLMs to SQL, NoSQL, and vector databases.
Database MCP servers allow AI agents to query schemas, read records, and write data through a natural-language interface. Use them for analytics, reporting, and data-driven agents.
Top picks
TypeDB MCP Server
Enables AI assistants to interact with TypeDB databases, allowing LLMs to execute TypeQL queries, manage databases, and manage users through natural language.
doctree-mcp
Enables agentic document retrieval over markdown, CSV, and JSONL using BM25 and tree navigation, without vector databases or embeddings, allowing AI agents to search, browse, and retrieve structured document sections.
hive-memory
MCP server for personal and shared memory with full-text search (SQLite FTS5) and outcome tracking, compatible with any MCP agent like Claude Code, Cursor, and Codex CLI.
io.github.oxgeneral/agentmem
Lightweight persistent memory for AI agents using a single SQLite file with hybrid search (keywords + semantics). Zero to 12MB install, no cloud or server required.
KIAgent
Personal digital memory: a desktop app that indexes your mail, chats, and documents into a local SQLite corpus on your machine — parsing and OCR run on-device via llama.cpp — and serves them to any MCP client. Local server on 127.0.0.1, plus an optional per-user remote endpoint (you.localkiagent.com) whose TLS terminates on your own device.
Fusion Schema MCP Server
Grounds a Fusion-SQL agent against Oracle Fusion schema catalog, enabling table/column search, validation, and relationship discovery.
How to choose
- Prioritize transport: stdio servers run locally; HTTP/remote servers need no install but require network trust.
- Check stars & maintenance: Higher GitHub stars and recent commits usually signal a more reliable project.
- Review security scope: File-system and browser servers can access sensitive data; run them in isolated environments.
- Match your client: Not all MCP clients support every transport; verify compatibility with Claude Desktop, Cursor, or your own agent.
Editorial picks are based on public metadata and keyword matching. They do not constitute endorsements; always review source and license before use.