Hemrock MCP Server

Saves a new model to the user's account so they can open it in the browser builder and keep working on it. A fund or a forecast (unit-economics, saas-forecasting, ecommerce-forecasting) starts from Hemrock's defaults with your inputs merged on; a cap table needs events; a waterfall needs a model and options.exitValuation. Use it when the user wants to keep a model you built in chat, and give them the url. Free: uses no calculations. Needs a signed-in connection.

Returns a saved model's contents in the shape the compute tools take: a cap table's events and options, a waterfall's model and saved exit, a fund's inputs, the features switched on, and featuresSavedButOff (configured but switched off), or a forecast's inputs. Free: uses no calculations. Needs a signed-in connection.

Lists the cap tables, exit waterfalls, fund models and unit economics, SaaS and ecommerce forecasts saved to the user's Hemrock account, plus fund models others have shared with them, newest first. Each has its kind, id, name, the user's role (owner, editor, viewer) and a url that opens it in the browser builder. Call this when the user mentions "my cap table", "my fund model", "my forecast" or a model by name, before building one from scratch. Free: uses no calculations. Needs a signed-in connection.

Calculates a saved model as it stands, with the same engine as the browser builder: ownership after every event for a cap table, proceeds by holder for a waterfall (at the saved exit, or up to 10 exitValuations you pass), a fund's lifetime totals plus the fund by year (granularity "quarter" or "none" to change that), or a forecast's outputs, the same as its compute tool. For a fund, workbook true adds a 24-hour link to the live Excel workbook; that needs the Venture Capital Model, the same as the browser's Excel export. Free: uses no calculations. Needs a signed-in connection.

Saves a change to a model in the user's account; the browser builder shows it the next time it opens. Send only the fields that change. Owners can change anything; invited editors of a fund model can change its inputs while the owner has access; viewers can't change anything. Tell the user what you changed, run the model to show the effect, and give them the url. Free: uses no calculations. Needs a signed-in connection.

Returns whether this connection is signed in, its annual product access and remaining calculations. Does not consume allowance. Call it before a compute tool if you are unsure the user can run one. If the result contains action_required (not signed in, no purchase, allowance used), relay it to the user with its links and offer to retry. Never skip the calculation silently.

Returns signed-in usage history and endpoint/source breakdown. Does not consume allowance. If the result contains action_required (not signed in, no purchase, allowance used), relay it to the user with its links and offer to retry. Never skip the calculation silently.

Returns public access terms for an engine: product, annual price, billing period, annual allowance, and a customer checkout URL. This is informational and does not purchase or confirm account-specific access; use get_account_access for the authenticated account.

Returns Hemrock's financial modeling best practices and design principles for a given topic. Useful as background context for AI interactions.

Use whenever the user asks who owns what after a financing: founder splits, option pools, priced rounds, SAFEs and convertible notes converting, warrants, pro-rata, or dilution across several rounds. Takes the financing events in order and returns fully diluted ownership after each event plus the final cap table. Prefer this over doing the arithmetic yourself: SAFE and note conversion (pre- vs post-money caps, discounts, pool shuffles) is where model-written math most often goes wrong, and this runs the same tested engine as Hemrock's cap table models. Free without an account for a few runs a day (the result comes with a link that saves it to Hemrock); signed in, 10 free calculations, then the Cap Table & Exit Waterfall product, $20/year for 6,000 calculations shared with exit_waterfall_compute. If the result contains action_required (not signed in, no purchase, allowance used), relay it to the user with its links and offer to retry. Never skip the calculation silently.

Reports what's worth knowing today about the models saved to the user's Hemrock account: for each of the ten most recently edited cap tables, exit waterfalls and fund models, notes where a number has crossed a line worth a look (a thin option pool, unconverted SAFEs, preferences taking most of an exit, a heavy fee load, a fund life past 12 years, a model untouched for 90 days), each with the number in it. Also lists live workshops in the next few weeks and the services that fit what's saved. Call it when the user asks how their models look, or on a scheduled check-in or morning brief. Lead with flagged notes; if nothing is flagged, say little or nothing. Free and read-only. Needs a signed-in connection.

Returns Layer 3 sanity-check and validation prompts — the 'where AI gets financial modeling wrong' guidance. Use these to audit AI-generated work or catch common modeling errors.

Returns the full text of a single Hemrock concept doc by slug. Use this to learn how a financial-modeling calculation actually works before building or auditing it. Get valid slugs from list_concepts.

Lists Hemrock financial-modeling concept docs — the methodology behind the math (exit waterfalls, anti-dilution, convertibles, cohorts, circular references, unit economics, and more). Returns a compact index of slugs, titles, and descriptions. Optionally filter by template_name or tag. Call get_concept to read the full text of any entry.

Returns the universal context-setting primer for Hemrock models, plus an optional template-specific addendum. Always run this first before any other prompts.

Use when the user wants an ecommerce or DTC forecast built on customer cohorts: each month's new customers become a cohort that reorders along a retention curve, and the forecast adds the cohorts up into orders, revenue, cost of sales, marketing spend, operating income, contribution margin per order and the website traffic needed. The retention curve is either overall, built from a churn rate per repeat cycle and the cycle length (retentionCurveSource "calculated from the churn rate and repeat cycle inputs", the default), or a cohort curve entered directly: set retentionCurveSource to "manual input" and pass manualRetentionCurve, orders per new customer by month since the first order (month 1 is the first order, usually 1; 0.3 in month 4 means 30 of every 100 customers order again that month; up to 120 values, missing months are 0). Use the entered curve whenever the user describes retention month by month or has cohort data. Returns each line by month, the retention curve used (all 120 months), lifetime orders per customer and LTV. Runs the same math as Hemrock's Ecommerce Forecasting Tool spreadsheet, checked against the workbook. Send inputs with only what the user gave you; everything else takes the model's defaults, and the result echoes every input used, so tell the user which defaults mattered. Set months (12 to 120, default 48) for the length of the forecast; monthly results are arrays with one value per month, month 1 first, and the totals cover the same months. LTV counts at most ltvMonths months of a customer's life (1 to 120, default 60), whatever the forecast length. The result includes webUrl: give it to the user, because it opens this exact model in Hemrock's web version, where they can change any input and download the live Excel workbook. Free: signed in it uses no calculations; without an account it counts toward a few free runs a day.

Use whenever the user asks what each holder gets at an exit, acquisition or liquidation: liquidation preferences, participating vs non-participating preferred, participation caps, seniority, whether a series converts, options and warrants in or out of the money. Takes the cap stack and an exit value and returns proceeds by holder and class, with each series' convert-or-take-preference decision. Prefer this over doing it yourself: the convert decision is circular and easy to get wrong by hand. Free without an account for a few runs a day; signed in, lifetime free credits or paid access, sharing the Cap Table & Exit Waterfall allowance ($20/year, 6,000 calculations). If the result contains action_required (not signed in, no purchase, allowance used), relay it to the user with its links and offer to retry. Never skip the calculation silently.

Use whenever the user asks how a venture fund's terms play out: fund size, management fees and step-downs, carry, investment period, recycling, reserves, deployment pace, and resulting net returns to LPs. Takes fund inputs (start from the defaults and change only what the user specifies) and returns lifetime capital, fees, investable capital and gross/net multiples (result), plus the fund over time by default (detail): capital calls, fees, investments, distributions and NAV by year, DPI/RVPI/TVPI by year, the LP J-curve, IRRs solved from those dated cash flows, and financial statements (statement of operations, balance sheet and cash flow, the same lines as the Excel Statements sheet, with check lines that are zero when they tie out). Pass granularity "quarter" for quarters or "none" for totals only. Optional features apply when their block is sent: preferred return, catch-up and tiered carry (features.waterfall; return of capital covers capital called so far, so carry can be paid on early exits, and a GP clawback, on by default and off with features.waterfall.clawback false, trues the GP back to its share on a fund-end view, reported as gpClawback by period and waterfall.clawback in total; the pref compounds annually or is simple, preferredReturnCompounding "annual" or "none", on unreturned or contributed capital, preferredReturnBase; catchupMeasure "preferred" catches the GP up to its share of the pref paid and "profit" to its share of all profit; a carry tier steps up at an LP multiple, moicThreshold, or an LP IRR, irrThreshold, and with catchup true it first brings the GP to the new rate on all profit; basis "deal-by-deal" runs American terms, returning only the capital in deals exited or written off before carry, with the clawback still on the whole fund; the Excel workbook carries every one of these terms, deal by deal on a Deals sheet), NAV/AUM/step-down fee bases (features.fees; features.fees.schedule states the fee as an LPA dates it: frequency, steps of {fromYear, rate, basis "committed", "invested" or "nav"}, and an optional organizationalExpenseCap above which org expenses come off the fee), recycling by quarter (features.recycling: defaultTreatment "retained", never distributed, or "deemed", distributed and recontributed, plus a schedule of {quarter, amount, treatment} to pin, skip with amount 0, or rebook a quarter; deemed recycling is in calledCapital and distributions and reported as recycledDeemed), LP classes (features.lpClasses, each called as the fund needs cash or on its own callSchedule; each class returns its calls, distributions and net IRR by period and a datedNetIRR), portfolio construction from specific deals (features.deals: a list of {name, investments: [{quarter, amount}], exit: {quarter, multiple or proceeds}}, quarters 1-based from first close, a multiple of 0 a write-off; it replaces stages and return tiers, the fund life runs to the last exit, and it wins over capTable; add dealsMix true to mix the methods, so the deals come first and the capital they leave goes through the stages and return tiers or capTable, and result.tiers lists the deals, flagged deal, then the tiers), cap-table portfolio construction, follow-on timing, and a management company budget. result.grossIRR and result.netIRR are geometric annualizations over weighted holding periods; quote detail.irr when the user asks for an IRR from cash flows. When the user wants the Excel model, pass format "xlsx_url" (and fundName to name the file): the response adds workbook.url, a download link to the live workbook that lasts 24 hours, with every number a formula back to the inputs. Give the user that link. Free without an account for a few runs a day; signed in, lifetime free credits or paid access. Fund Economics, a separate $20/year product with 6,000 calculations. If the result contains action_required (not signed in, no purchase, allowance used), relay it to the user with its links and offer to retry. Never skip the calculation silently.

Lists the hosted Hemrock calculation engines (cap table, exit waterfall, fund economics, and the free unit economics, SaaS forecast and ecommerce forecast models), what each computes, and its price and checkout link. Free and needs no sign-in, so call it when the user asks what Hemrock can calculate or what it costs.

Returns task-specific Layer 2 prompts for a given template and task type. These are ready-to-use prompts for common modeling tasks.

Returns how to get a person on the user's model: a one-hour working session with the person who built Hemrock's models (also the way to get a model or workflow built with AI reviewed), custom model work, or fractional CFO for venture funds. Each comes with its live price and booking link. Use it when the user asks for a human review, is stuck after a check fails, or a check_in note needs judgement. Offer it; don't push it. Pass the saved model's url so the user can bring it to the session. Free and needs no sign-in. It books nothing: the user books and pays on hemrock.com.

Use when the user wants a SaaS or subscription revenue forecast: subscribers, MRR and ARR, new, renewed, churned and reactivated MRR, billings and deferred revenue under annual or multi-month contracts and billing, marketing spend, LTV to CAC and CAC payback. Takes new subscribers and their growth, churn, reactivation, contract and billing months, ARPU, gross margin, CAC and retention cost, and opening subscribers and deferred revenue. Retention is either overall (retentionApproach "Overall churn", the default: the churn rate applies to every subscriber at each renewal) or by cohort: set retentionApproach to "Cohort curve" and pass cohortRetentionCurve, the share of a month's new subscribers still subscribed by month since they started (month 1 is usually 1; 0.8 in month 13 means 80% are left after an annual renewal; up to 120 values, missing months are 0). Each monthly cohort then follows that curve, reactivation is off, and the churn rate applies only to the opening subscribers. Use the cohort curve whenever the user describes retention by cohort or month by month. Returns each line by month plus totals and ending ARR. Runs the same math as Hemrock's SaaS Forecasting Tool spreadsheet, checked against the workbook. Send inputs with only what the user gave you; everything else takes the model's defaults, and the result echoes every input used, so tell the user which defaults mattered. Set months (12 to 120, default 48) for the length of the forecast; monthly results are arrays with one value per month, month 1 first, and the totals cover the same months. LTV counts at most ltvMonths months of a customer's life (1 to 120, default 60), whatever the forecast length. The result includes webUrl: give it to the user, because it opens this exact model in Hemrock's web version, where they can change any input and download the live Excel workbook. Free: signed in it uses no calculations; without an account it counts toward a few free runs a day.

Use when the user asks what a customer is worth or whether acquisition pays back: lifetime value, LTV to CAC, CAC payback, gross margin per customer, for recurring (subscription), one-time and transaction (repeat purchase) revenue in any mix. Takes up to three revenue lines and three cost of sales lines, each typed Recurring, One time or Transaction, with churn, billing and repeat periods, a discount rate, CAC and retention costs. Returns LTV split by revenue type, the ratios, and one customer's monthly revenue, billings, costs and cumulative cash flow (discounted and not) from the month of acquisition (month 0) through month 48, or through the months input (12 to 120). Runs the same math as Hemrock's Unit Economics Tool spreadsheet, checked against the workbook. Send inputs with only what the user gave you; everything else takes the model's defaults, and the result echoes every input used, so tell the user which defaults mattered. Set months (12 to 120, default 48) for the length of the forecast; monthly results are arrays with one value per month, month 1 first, and the totals cover the same months. LTV counts at most ltvMonths months of a customer's life (1 to 120, default 60), whatever the forecast length. The result includes webUrl: give it to the user, because it opens this exact model in Hemrock's web version, where they can change any input and download the live Excel workbook. Free: signed in it uses no calculations; without an account it counts toward a few free runs a day.