Liquid
Auto-discover and query APIs, databases, and other agents from one MCP bridge with optional read-only or write execution.
Overview
Liquid MCP is a MCP server for the Build phase that auto-discovers and connects your agent to APIs, databases, and other agents with read-first and optional write execution.
What is this MCP server?
- Connect agents to APIs, databases, and other agents with auto-discovery
- Read-first default; set LIQUID_ALLOW_WRITES=1 to expose liquid_execute writes
- PyPI package liquid-mcp with uvx runtime hint
- Supports OpenAI, Gemini, or Anthropic keys for discovery (plus compatible base URLs)
- Single MCP surface instead of one-off server per integration
- PyPI identifier liquid-mcp; server version 0.2.0 in manifest
- Write tool gated by LIQUID_ALLOW_WRITES environment flag
- stdio transport with uvx runtime hint
What problem does it solve?
Maintaining a separate MCP server for every API and database burns solo-builder time and fragments agent context.
Who is it for?
Builders who want a generic integration MCP during feature work and occasional ops reads without authoring dozens of custom tool manifests.
Skip if: Teams that need audited, single-purpose connectors only, or who cannot supply an LLM API key for discovery.
What do I get? / Deliverables
One Liquid connection gives your agent discovered read tools—and controlled writes when enabled—across heterogeneous backends.
- Auto-discovered MCP tools against configured APIs and data sources
- Read-only agent access by default for safer exploration
- Optional write execution path when LIQUID_ALLOW_WRITES is enabled
Recommended MCP Servers
Journey fit
Universal connectors land on the build shelf because wiring external systems is the primary job; ops and analytics reuse the same bridge. Liquid is integration infrastructure—discovery and execute tools—not frontend or docs work.
How it compares
Generic auto-discovery integration MCP, not a curated benchmark feed or dependency security oracle.
Common Questions / FAQ
Who is Liquid MCP for?
Indie developers and agent-heavy teams who integrate many external systems and want discovery-driven tools instead of hand-built MCP per service.
When should I use Liquid MCP?
During build when wiring backends and third-party APIs, and optionally later for read-only operational or analytics queries from the same agent setup.
How do I add Liquid MCP to my agent?
Run liquid-mcp from PyPI via uvx or your Python workflow with stdio transport, set at least one discovery key such as OPENAI_API_KEY (or Gemini/Anthropic alternatives), and register the server in your MCP client; set LIQUID_ALLOW_WRITES=1 only if you need write tools.