SLayer is a MIT-licensed open-source semantic layer maintained by the Motley team, focused on giving AI agents governed data access and querying. It supports MCP among several interfaces and has clear install steps, making it well suited to teams building analytical agents or APIs on a warehouse; its read-only connections and credential-passing approach warrant attention to security configuration.
SLayer is an embeddable, expressive semantic layer created and maintained by the Motley team, enabling AI-powered data analytics on top of your warehouse. It gives agents a governed, shared surface through which they access your data and metrics and deliver reliable answers. SLayer handles database connectivity (read-only), SQL translation, common data transformations, and row-level security, so LLMs and humans don't have to. It can be used as a standalone tool or imported as a Python library and embedded into any Python app — for powering analytical MCP servers or APIs, or simply querying databases semantically. Unlike traditional BI semantic layers where every metric and aggregation must be predefined, SLayer lets you define a column such as revenue once and query it using expressions like revenue:sum, revenue:avg, revenue:sum / *:count, or time_shift(revenue:sum, -1, 'year'), with multi-stage queries also supported. SLayer focuses on the common agentic search → inspect → query flow, offering a search tool for efficient discovery and a memory store for linking relevant business context. Agents, apps, and humans can talk to SLayer via MCP, REST API, CLI, Python client, Flight SQL, or a Postgres-based SQL API, and it provides importers for dbt, Cube, and Ossie configs.
Using uv is recommended. Run uv tool install 'motley-slayer[all]'; if slayer isn't found on PATH afterwards, run uv tool update-shell and reopen your terminal. With the demo dataset: claude mcp add slayer_demo -- slayer mcp --demo (preloaded Jaffle Shop demo, zero-config). With your own data: create a datasource, e.g. slayer datasources create 'postgresql://user:${DB_PASSWORD}@hostname/db_name' — the password is read at init time and is not saved to disk or exposed to Claude — then run claude mcp add slayer -- slayer mcp --ingest-on-startup. Launch Claude Code from a shell where DB_PASSWORD is exported, since the MCP subprocess inherits its environment from the launching process.