OpenTax by Invaro

OpenTax by Invaro is a deterministic tax computation engine designed to turn any AI agent into a precise tax preparer. Language models operate probabilistically while tax law requires exact outcomes, creating a mismatch that OpenTax addresses by producing identical tax returns from the same facts every time with each line traceable to the originating statute.

It functions as an open-source engine under the AGPL-3.0 license and is described as the only open-source tax engine available. Every rule and test case is public. The engine delivers 96 percent exact returns on TaxCalcBench, an industry benchmark that evaluates preparation of 50 complete US tax returns where every line must match precisely. This score represents the highest recorded on the benchmark. In comparison, Claude Sonnet alone achieves 6 percent exact returns on the same test. Connecting the model to OpenTax via a single MCP URL raises performance from 6 percent to 96 percent, showing that the engine sets the effective ceiling rather than the underlying model.

The tool emphasizes rules-based exact arithmetic instead of sampling, eliminating any possibility of LLM drift. Installation takes 30 seconds through its GitHub repository. It connects to AI agents through one MCP URL in approximately two minutes with no additional configuration required. The engine amplifies the accuracy of any model pointed at it.