Boyce
Stop your agent from guessing SQL—compile natural-language or structured filters into deterministic SQL against PostgreSQL or Redshift.
Overview
Boyce is an MCP server for the Build phase that compiles StructuredFilter inputs into deterministic SQL for PostgreSQL and Redshift so agents stop guessing queries.
What is this MCP server?
- Deterministic SQL compiler: agents submit StructuredFilter plans instead of raw guessed SQL
- Optional live PostgreSQL or Redshift via BOYCE_DB_URL for profiling and EXPLAIN pre-flight
- stdio PyPI package (boyce)—typical local MCP install for Claude Desktop, Cursor, Claude Code, Windsurf
- Host model can do NL reasoning; Boyce receives pre-built StructuredFilter when the MCP host supplies its own LLM
- BOYCE_PROVIDER and ANTHROPIC_API_KEY only needed when Boyce runs its own planner, not for most MCP hosts
- PyPI package identifier boyce at version 0.1.1
- Supports PostgreSQL and Redshift via BOYCE_DB_URL
- stdio transport for local MCP integration
Community signal: 1 GitHub stars.
What problem does it solve?
Agents that write SQL from scratch routinely hallucinate columns, dialect syntax, and costly query plans.
Who is it for?
Indie builders connecting coding agents to Postgres or Redshift who want a compiler-style SQL layer instead of pasted NL-generated queries.
Skip if: Projects with no SQL warehouse, or teams that already enforce SQL only through reviewed migrations and forbid agent-generated queries entirely.
What do I get? / Deliverables
After registering Boyce, your agent can produce compiled SQL from structured filters with optional live profiling and EXPLAIN pre-flight when BOYCE_DB_URL is set.
- Deterministic SQL output from agent-supplied StructuredFilter plans
- Optional schema-aware profiling and EXPLAIN pre-flight against configured databases
- Reduced reliance on raw LLM-generated SQL in agent workflows
Recommended MCP Servers
Journey fit
Query generation and schema-aware SQL belong in the build phase when you wire backend data access and analytics features. Boyce sits on the database/backend path: profiling, EXPLAIN pre-flight, and StructuredFilter→SQL compilation for agent-driven apps.
How it compares
Deterministic SQL compiler MCP server—not a natural-language BI dashboard skill or a generic database browser.
Common Questions / FAQ
Who is Boyce for?
Solo builders and agent developers who query PostgreSQL or Redshift through MCP and want structured, deterministic SQL compilation instead of ad hoc model output.
When should I use Boyce?
Use it while building backend or analytics features where the agent must translate filters or intents into SQL, especially when you want EXPLAIN pre-flight against a live BOYCE_DB_URL database.
How do I add Boyce to my agent?
Install the boyce PyPI package and configure stdio MCP in your client; set BOYCE_DB_URL for live profiling. Most hosts use their own model for NL→StructuredFilter and do not need BOYCE_PROVIDER or ANTHROPIC_API_KEY.