Tool reference

The hosted Breakpoint Engine service serves 15 tools over MCP Streamable HTTP: opm_forwardsolve, opm_backsolve, the nine opm_scenario_* operations, plus opm_estimate, opm_explain_breakpoints, opm_cancel, and opm_schema. Work units are authorized and settled against your account balance.

Solves

opm_forwardsolve — classic OPM forwardsolve (emit_detail optional).
opm_backsolve — solve for equity value or volatility.

Discovery, estimation, and control

opm_schema — discovery metadata; call first in a new session.
opm_estimate — work-unit estimate before a metered solve. Returns work_expected (the predicted cost) inside a work_min / work_max band, plus forwardsolves_expected. The gateway sizes and prices the solve on its own when you run it.
opm_explain_breakpoints — the breakpoint schedule on its own, ahead of a priced solve.
opm_cancel — stop an in-flight job by job_id; it returns the partial schedule and event log produced so far, and that delivered work settles against your balance.

Interactive scenarios

opm_scenario_begin — start an ephemeral txn (pause after schedule build).
opm_scenario_step / opm_scenario_query — advance and inspect frames.
opm_scenario_mutate / opm_scenario_rewind — change terms and replay.
opm_scenario_branch / opm_scenario_diff — compare alternatives.
opm_scenario_commit / opm_scenario_abort — finish or discard; the txn lives in-process for the session.

Resources

opm://schema/cap_table, opm://schema/assumptions
opm://methodology/overview, opm://protocol/version

Methodology vintage

assumptions.methodology selects the valuation-guidance vintage a solve runs under, and the vintage is echoed in every result rather than inferred.

"2011" — the default. Absent the field, this is what you get: the methodology the engine inherited from the predecessor implementation it was ported from.
"2026" — a newer vintage informed by the working draft of an updated AICPA Accounting and Valuation Guide for privately held company equity securities, issued 2025-12-18 by the Equity Securities Task Force with FinREC (public comment closed 2026-06-01). It is a working draft, not the final published guide, and this vintage continues to be developed as the draft is worked through.

The vintages do not always agree — where the draft changes a rule, "2026" follows the draft and the default does not. Some capabilities are gated on it (discount_model: "ghaidarov" requires "2026"). Reproducibility holds within an engine version: the same inputs on the same version return the same numbers.

Conventions

Money and rates as decimal strings.
Recommended flow is opm_estimate → execute. The estimate quotes work_expected / work_min / work_max; the gateway sizes the job itself, admits it when your balance covers max(work_min, floor) and settles actual consumption capped at its own work_max quote. For a backsolve, pass the same target, tolerance, target_value, and max_iterations to the estimate that you will pass to the solve — the estimator models the solver's search from them, and defaults widen the band.
Seed equity_value near the answer you expect on a backsolve. The result is identical for any seed — the seed sets what reaching it costs. The solver searches [seed/4, seed×4], so any seed within 4× is free; each further 4× of distance buys another nested forwardsolve. Use the post-money valuation, or target_value ÷ that security's expected ownership. A placeholder seed against a $3.125M target measured 18% more work than a realistic one, and opm_estimate shows that before you spend — if work_expected looks high, re-seed and estimate again.
402 / insufficient_credits → buy a pack on account.

Input shape: preparing capitalization inputs. Connect: Cursor & VS Code, Claude, ChatGPT.