Anyquery: query GitHub, Notion and dozens more apps with SQL

An agent that needs GitHub issues, a Notion database and a local CSV usually needs three MCP servers, each with its own tools and parameters. Anyquery offers one interface instead. It’s a single-binary query engine built on SQLite that uses plugins to map apps, files and databases to SQL tables. If the agent can write SQL, it can query across sources and even JOIN them.

What it does

• Three MCP tools: listTables lists available tables, describeTable returns columns, supported operations and SQL examples, and executeQuery runs the query.
• Sources: plugins cover GitHub, Notion, Apple Notes, Chrome, Todoist, Spotify and more; it also queries local files and connects to PostgreSQL, MySQL and SQLite.
• Not just reads: tables that support writes accept INSERT, UPDATE and DELETE, and describeTable says which operations each table allows.
• Other modes: it can run as a MySQL server for TablePlus or Metabase, and anyquery gpt covers function-calling clients like ChatGPT that don’t speak MCP.

Who it’s for

• Individuals who want answers to cross-app questions like “what did I finish in GitHub and Todoist this week?”
• Analysts and developers comfortable with SQL who’d rather use one tool for many sources than install a stack of MCP servers.

Setup

Homebrew on macOS; APT, YUM, Scoop and Winget packages elsewhere. Then start it over stdio from your client config:

brew install anyquery
anyquery mcp --stdio

Install and authorize a plugin for each app you need, following the integration guides on the website.

Our take

About 1,800 stars as of 2026-10-06. Collapsing “connect lots of services” into one SQL interface is the most distinctive idea in this category: the tool count stays at three no matter how many sources you add, which saves context, and models are generally good at SQL. The downsides: plugin quality depends on each plugin’s author, so coverage is shallower than vendors’ own MCP servers, and executeQuery can write, so be careful pointing it at apps where the agent could delete data. The top of the README carries an ad for an LLM API reseller unrelated to the project. The core engine is AGPL-3.0; plugins carry their own licenses.