OunceMCP 1.0 RC: Token Budgets and Safe Writes for MCP Servers, 58-85% Fewer Tokens
gentleGiant96 · reddit · 2026-11-17
A developer released OunceMCP, a Python framework on FastMCP that makes careful MCP server behavior the default: per-call result token budgets with schema-preserving trimming and pagination, CI checks on tool definition size, automatic search-then-call for large catalogs, idempotent/approvable writes, and tenant-partitioned security defaults.
Benchmarks across 4 models and 840 runs: 58–85% fewer tokens per task on large-result servers, and 100% vs 68–70% task success on write scenarios where every model on plain FastMCP performed unconfirmed or unauthorized deletions. Trade-off: tool definitions grow 1.8–2.4x, so small-result servers cost more tokens. Apache-2.0, with a 15-chapter tutorial book and an audit tool for existing MCP servers.
More from coding & agent
- agent-browser: open-source browser automation so agents can verify fixes and file PRs — evilrabbit_ · 2026-11-17
- OpenRouter compares the context-token cost of MCP, CLI and Skills — rseroter · 2026-11-17
- PSN: storing agent skills as executable programs in an evolvable network — burkov · 2026-11-17
- Together built its own agent evals, lifting agent task success from 67% to 96% — togethercompute · 2026-11-17
- Solo Dev Shares 6 Hard-Won Lessons on Prompting Video With a Real Person's Face — pvdyck · 2026-11-17
- Who Can Even Read AI-Generated Code? Garbled Screenshot Sparks Debate — gethackteam · 2026-11-17