Anyquery MCP Server — SQL Everything: Query 40+ Apps, Files, and Databases from Your AI Agent

Part of our Databases MCP category.

At a glance: ~1,600 GitHub stars · v0.4.4 (October 2, 2025, no release in 7 months) · Go · AGPLv3 license · 47 plugins + 13 database connectors + 22 SQLite extensions · stdio/SSE/HTTP transports

Most MCP servers connect to one thing. The Anyquery MCP server by Julien Cagniart connects to everything — or tries to. Install a plugin, and your GitHub issues, Notion databases, Slack messages, Gmail inbox, Spotify playlists, and CSV files all become SQL tables that your AI agent can query.

That breadth is the pitch. One binary, one SQL interface, 40+ data sources. No per-API MCP server to install and configure separately. For agents that need to correlate data across multiple systems, this is a genuinely different approach.

What It Does

Anyquery is a SQL query engine built on SQLite. When you run anyquery mcp --stdio, it becomes an MCP server that exposes your installed plugins as queryable tables. Your AI agent discovers the available tables, understands their schemas, and writes SQL to answer questions.

The core workflow:

1. Install plugins for the apps you use (anyquery install github, anyquery install notion, etc.)
2. Configure credentials for each plugin (API tokens, OAuth, etc.)
3. Start the MCP server — your agent can now query across all sources with SQL

Plugin Ecosystem

The 47 available plugins span SaaS apps, local applications, and developer tools (down from 54 at launch — some consolidation since the October 2025 release):

Beyond plugins, Anyquery connects directly to 13 databases: MySQL, PostgreSQL, SQLite, DuckDB, ClickHouse, Cassandra, MariaDB, CockroachDB, ScyllaDB, SingleStore, TiDB, YugabyteDB, and AWS Aurora (MySQL and PostgreSQL variants).

The Power of SQL Joins

The real value isn’t querying one source — it’s joining across sources. An agent can write:

SELECT g.title, n.status
FROM github_issues g
JOIN notion_tasks n ON g.title = n.name
WHERE g.state = 'open' AND n.status != 'Done'

Find GitHub issues that don’t have matching Notion tasks marked done. Try doing that with separate MCP servers for GitHub and Notion — the agent would need to fetch from both, then correlate in context. With Anyquery, it’s one SQL query.

Three Ways to Connect

Anyquery offers three transport modes for MCP:

The MySQL server mode (port 8070, MySQL 5.7-compatible) isn’t MCP — it’s a separate capability that lets you connect any MySQL client to your Anyquery data. Useful for visualization tools and dashboards.

Setup

Claude Desktop

{
  "mcpServers": {
    "anyquery": {
      "command": "anyquery",
      "args": ["mcp", "--stdio"]
    }
  }
}

Claude Code

claude mcp add anyquery -- anyquery mcp --stdio

Before starting, install and configure your plugins:

# Install Anyquery
brew install julien040/anyquery/anyquery

# Install plugins
anyquery install github
anyquery install notion
anyquery install google_sheets

# Each plugin prompts for credentials (API tokens, OAuth)

Plugin configuration happens once. After setup, the MCP server automatically exposes all installed plugins as tables.

Performance

According to available documentation and analysis:

• Sub-second latency for most queries via on-demand API fetching
• 60-second response caching reduces redundant API calls
• Predicate pushdown filters data at source APIs where supported, reducing data transfer
• Memory footprint under 50MB — lightweight for what it does
• Read-only — write operations are in development but not yet available

The read-only limitation matters. Your agent can query GitHub issues but can’t close them. Can read Notion pages but can’t update them. For data analysis and correlation this is fine; for workflow automation, you’ll still need separate write-capable MCP servers.

Security Considerations

Anyquery handles sensitive credentials (API tokens for GitHub, Notion, Gmail, etc.), so security design matters:

• Token encryption via anyquery config set --encrypt — encrypts stored API credentials
• Localhost binding recommended for server mode — don’t expose the MySQL port to the network
• Schema-only transit — only table schemas pass through the MCP protocol; raw credentials stay on-device
• CVE-2025-61679 was patched in v0.4.4 — the fix added default authorization for MCP HTTP servers
• AGPLv3 license means modifications to Anyquery itself must be shared (plugins are independently licensed)

The v0.4.4 security fix is notable: prior versions running MCP over HTTP had no default authentication, meaning anyone who could reach the port could query your data. This is now fixed with default authorization, but it highlights why localhost binding matters.

Alternatives and Comparison

How does Anyquery compare to building a stack of individual MCP servers?

The trade-off is clear: Anyquery gives you breadth (many sources, one interface) while individual servers give you depth (full API coverage per source). The GitHub MCP server from Anthropic exposes 30+ tools covering issues, PRs, repos, branches, and code search. Anyquery’s GitHub plugin exposes tables you can query — powerful for reads, but no write operations and fewer specialized capabilities.

DBHub is the closest alternative — a universal database MCP server focused on traditional databases (MySQL, PostgreSQL, SQLite, MongoDB, etc.) with text-to-SQL capability. But DBHub doesn’t have the SaaS plugin ecosystem that makes Anyquery unique.

Who Should Use This

Good fit:

• Teams querying data across multiple SaaS tools (the cross-source SQL joins are genuinely useful)
• Data analysts who want BI-style queries over operational data without building a data warehouse
• Developers who want one MCP server instead of managing ten separate ones
• Anyone doing DevOps correlation (GitHub + monitoring + infrastructure)

Not ideal for:

• Agents that need to write/update data (read-only limitation, no write support after 7+ months of development)
• Projects needing deep API coverage for a single service (individual MCP servers go deeper)
• Organizations with AGPL licensing concerns (core engine is copyleft)
• Production systems needing guaranteed uptime (depends on plugin quality and source API stability)
• Teams that need active maintenance and bug fixes (no releases since October 2025)

The Licensing Question

AGPLv3 is polarizing. If you modify the Anyquery core engine and deploy it as a service, you must share your modifications. This doesn’t affect plugin development (plugins are independently licensed) or normal usage (running Anyquery as a tool), but it can be a dealbreaker for companies with strict open-source policies.

The plugins themselves are not subject to AGPL. Each plugin has its own license, and the RPC library connecting plugins to the engine is MIT-licensed — deliberately permissive to encourage ecosystem growth.

Rating: 3.5 / 5

(Downgraded from 4.0/5 in the May 2026 refresh — see below.)

Anyquery occupies a unique position in the MCP ecosystem: the universal data bridge. No other single MCP server connects to this many data sources through a unified SQL interface. The cross-source JOIN capability alone justifies installation for teams working across multiple tools.

What changed since March 2026: No new releases in 7 months (v0.4.4 remains current as of May 2026). The plugin count declined from 54 to 47 — some integrations were consolidated or removed. Write support, promised for development, has not shipped. The project now shows signs of extended dormancy rather than active iteration.

The rating drops to 3.5/5. The core concept remains strong and the tool still works, but the stalled development cadence and shrinking plugin ecosystem are meaningful negatives for teams evaluating production dependencies. The original 4.0/5 assumed active development would close the write-operations gap. After 7 months without a release, that assumption no longer holds.

For agents that need to answer questions across multiple data sources — “which GitHub issues assigned to me don’t have Notion tasks?” or “show me all Salesforce deals alongside their Slack channel activity” — Anyquery is still the most efficient read-only path available today. But if write operations are required, individual MCP servers per data source remain the only viable option.

This review was originally published March 23, 2026 and refreshed May 3, 2026. Refresh findings: v0.4.4 remains current (7 months since last release), plugin count declined 54→47, no write support added, project appears dormant. We research MCP servers thoroughly but do not test them hands-on. Visit the Anyquery GitHub repository for the latest information.