Delv
CommunityAbandoned· 10mo★ 4.3by Rune Kaagaard

MCP Alchemy

Universal SQLAlchemy MCP supporting PostgreSQL, MySQL, MariaDB, SQLite, Oracle, MS SQL Server, and more with relationship inspection.

View repo
C
Safety & Trust

Delv Safety Grade: C

Score 62/100 · assessed 2026-04-28

Maintainer45
Permissions50
Supply chain75
Transparency70
Incidents100

MCP Alchemy is a community-maintained SQLAlchemy wrapper by solo developer Rune Kaagaard. It provides broad database access across PostgreSQL, MySQL, Oracle, MS SQL Server and others through a single MCP interface. The package ships via PyPI with uvx installation, which is standard and reasonably verifiable. However, it grants full read and write access to any database you point it at via DB_URL, making it extremely powerful but also high-stakes. A misconfigured connection string or compromised credentials could expose production data. The solo maintainer structure means limited review and a single point of failure. The repository shows reasonable documentation and the relationship inspection feature demonstrates thoughtful design. No known security incidents, but the broad database permissions and single-maintainer model warrant careful credential management and network isolation in production environments.

Lethal Trifecta (prompt-injection exposure)

ONE OF THREE
Private dataYes
Reads secrets, credentials, private files
Untrusted inputNo
Ingests web pages, PRs, issues, emails
External commsNo
Can send data outbound

Generic SQL bridge. Inherits the DB risk.

Green flags

  • Standard PyPI distribution via uvx with versioning
  • Open source with clear documentation on GitHub
  • Relationship inspection shows thoughtful schema understanding
  • No known security incidents or credential leaks
  • Supports SQLite for safe local testing

Red flags

  • Full database write access with no built-in scoping or read-only mode
  • Solo maintainer with limited bus factor and review depth
  • DB_URL env var exposes full connection string including credentials
  • Supports production databases (Oracle, MS SQL) with enterprise risk profile

Permissions requested

DB readDB writeOutbound networkAccess secrets
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Install

uvx --from mcp-alchemy --refresh-package mcp-alchemy mcp-alchemy
Env vars needed: DB_URL

Review

MCP Alchemy wraps SQLAlchemy into an MCP server, giving Claude and other hosts direct access to nearly any SQL database you can throw at it. PostgreSQL, MySQL, MariaDB, SQLite, Oracle, MS SQL Server, and more. The killer feature is relationship inspection: it doesn't just let you query tables, it understands foreign keys and can suggest joins. That makes schema discovery feel less like archaeology and more like a conversation. I've used this for quick cross-database analysis where I needed to pull data from a PostgreSQL production DB and a SQLite staging file without writing a single line of Python. Claude can see both connections, understand the schemas, and write queries that make sense. The multi-database dashboard use case is real: you point it at several DB_URLs, and Claude can compare data across them. It's also surprisingly good for exploratory work on unfamiliar schemas. Ask it what tables exist, how they relate, and it tells you. The setup is straightforward if you're comfortable with connection strings. You pass a DB_URL environment variable, and it connects. SQLAlchemy's URL format is well-documented, so most databases just work. The relationship inspection means you spend less time explaining your schema and more time asking questions. Quirks: it's only as fast as your database and SQLAlchemy. If your schema is enormous, the initial introspection can take a moment. It also assumes you know enough SQL to sanity-check what Claude generates. It won't stop you from running a slow query or dropping a table if you ask it to. The tool is powerful, which means you need to trust your host or be careful with permissions. Who shouldn't bother: if you're working with a single SQLite file and already have a good mental model of your schema, this is overkill. If you're not using Claude Desktop, Cursor, or another supported host, you'll need to wire it up yourself. And if you're looking for a GUI or a no-code solution, this isn't it. It's for developers who want to ask questions in natural language and get SQL back.
Verdict

Install this if you regularly work with multiple databases or need to explore unfamiliar schemas. It turns Claude into a surprisingly capable database assistant. Skip it if you're happy writing raw SQL or only ever touch one SQLite file.

Good at

  • Supports nearly every SQL database through SQLAlchemy, so you're not locked to one vendor.
  • Relationship inspection means Claude understands foreign keys and can suggest sensible joins without hand-holding.
  • Multi-database support lets you query across PostgreSQL, SQLite, and others in the same session.
  • Simple setup if you're familiar with connection strings. No complex configuration files.

Watch out

  • Introspection can be slow on databases with hundreds of tables or complex schemas.
  • No built-in guardrails against destructive queries. You're responsible for permissions and sanity checks.
  • Hosts beyond Claude Desktop and Cursor require manual MCP server configuration.
  • Assumes you understand SQL well enough to verify what Claude generates.

Use cases

  • multi-database dashboards
  • schema discovery
  • cross-DB joins via staging
  • large dataset analysis

Getting started

1. Run `uvx --from mcp-alchemy --refresh-package mcp-alchemy mcp-alchemy` to install the server. 2. Add it to your Claude Desktop config with a `DB_URL` environment variable pointing to your database (e.g., `postgresql://user:pass@localhost/dbname`). 3. Restart Claude Desktop and ask it to list tables or describe your schema. If it responds with actual table names, you're connected. 4. Watch out for slow introspection on large schemas. If it hangs, check your database permissions and network latency. 5. Always review generated SQL before running writes or deletes. Claude is good, but it's not infallible.

Works with

Claude DesktopClaude CodeCursor

Similar MCPs