MCP App Store
by KeenEthics
ProFormaEngine icon

ProFormaEngine MCP Server for Claude and ChatGPT

By ProFormaEngine

Overview

ProFormaEngine lets Claude build and maintain multi-year financial models (pro formas, rate studies, budgets and debt plans) that live in your account rather than in a chat. Describe a model and Claude creates its inputs, line items and plain-text formulas such as revenue[t-1] * (1 + growth), so every number can be traced. It can add tables and charts, model bonds and loans with principal and interest, group and tag line items, and copy a model to try a scenario, then compare the two and show which results changed and why. Every model also appears in the ProFormaEngine web app as a spreadsheet, with formulas, precedents and dependents, and downloads as an Excel workbook in which formulas stay formulas. What Claude can do with the connector: • Read: list models; read values, summaries, tables and charts; compare two models; export a model as JSON • Build and edit: create and copy models; add inputs, line items, formulas, tables, charts and groups; set values, tags, labels, notes and formats • Remove: reset models, clear values, and remove items, tables and charts. Claude asks you to confirm these first, and may leave permanently deleting a model to you in the web app The connector only works on models in your own account. ProFormaEngine is invite-only for now: join the waitlist at www.proformaengine.com.

How to connect

  1. Claude: Open the ProFormaEngine listing in the Claude connectors directory and choose Connect. Claude listing ProFormaEngine exposes 47 tools here.
  2. ChatGPT: Open the ProFormaEngine listing in the ChatGPT app directory and choose Connect. ChatGPT listing ProFormaEngine exposes 52 tools here.

Tools

add_chart

ChatGPT
Add a new chart. A chart is a named, reusable plot definition — like a table, it stores what to show, never values, and renders against the model's current state every time (get_chart; the web app draws it). Charts have their own namespace: a chart may share its name with a table or line item. series is an ordered list; each entry picks its source: - {"source": "item", "name": <line item>, "label": null, "type": null, "axis": "left"} — one line item's values. label defaults to the item's label. - {"source": "tag", "tag": <tag>, "type": null, "axis": "left"} — one series per line item currently carrying the tag (re-evaluated every render; an unused tag plots nothing). - {"source": "scalar", "name": <scalar>, "label": null, "type": null, "axis": "left"} — the scalar's value as a flat reference line (e.g. a covenant minimum); drawn as a line unless type says otherwise. chart_type ("line" or "bar") is every series' default; a series can set its own type to mix bars and lines. axis: "right" puts a series on a second y-axis, so dollars and ratios can share a chart. start_period/end_period (inclusive, optional; a year, or a label like "2027-03") limit the periods shown. value_format formats the left axis and right_axis_format the right; each defaults to the first series on that axis. Rejected, changing nothing, if an item/scalar series names something that doesn't exist, start_period is after end_period, or a format is an unknown preset. A chart that plots an item or scalar blocks removing it; renaming one updates the chart.

add_groups

ChatGPT
Add one or more groups — named sections ("Assumptions", "Operating Revenue", "Debt Service") that line items and scalars can be filed under, so whole-model views (the spreadsheet view, the Excel export, get_model_summary) lay the model out in order instead of as one flat list. Appended after any existing groups, in the order given; reorder with reorder_groups. Groups are display only — never read by a formula, so they can't change a value. They are not tags: a tag is a calculation concept (tag["x"][t] sums its members) and an item can carry many; an item is in at most one group. get_model_summary's groups lists the existing ones, so a near-duplicate (revenue vs operating_revenue) can be avoided. Each entry: name (lowercase slug), optional label (the section heading people see; the name is shown without one) and notes. Then file items into them with set_group, or pass group when adding items. Each group is added independently: a bad entry is reported per item and doesn't block the rest. Returns compact JSON: one result per input item, in the same order: {"name": ..., "success": bool, "error": str | None}.

add_line_item

ChatGPT
Add a new period-indexed line item. A period resolves from values first if present, otherwise from formula. Provide either or both (e.g. seed one period in values and let formula compound from there via name[t-1]).

add_line_items

ChatGPT
Add several line items in one call — for loading/translating a model whose structure is already fully known, rather than building one interactively. See add_line_item's formula argument for the full grammar. Each item is still added independently (one bad formula is reported as a per-item failure and doesn't block the rest), but unlike a plain loop, declaration order within the batch doesn't matter: an item referencing another item later in the same batch is resolved once that sibling is added, not rejected for being "declared too early." A formula referencing a scalar or line item that isn't already in the model and isn't in this batch fails immediately — add scalars and any prerequisite line items first. A dependency cycle confined to this batch fails as a group, naming everything still unresolved. Prefer add_line_item one at a time when building interactively — each call's own confirmation lets you catch a mistake immediately. A values period outside the model's periods is staged until set_periods adds it, same as add_line_item. Returns compact JSON: one result per input item, in the same order: {"name": ..., "success": bool, "error": str | None}, plus "staged_periods": [...] on a successful item that has staged values.

add_scalar

ChatGPT
Add a new scalar constant (not period-indexed), e.g. a tax or growth rate.

add_scalars

ChatGPT
Add several scalars in one call — for loading/translating a model whose structure is already fully known, rather than building one interactively. Each scalar is added independently: one bad entry (e.g. a duplicate or invalid name) is reported as a per-item failure and doesn't block the rest. Prefer add_scalar one at a time when building interactively — each call's own confirmation lets you catch a mistake immediately. Returns compact JSON: one result per input item, in the same order: {"name": ..., "success": bool, "error": str | None}.

add_table

ChatGPT
Add a new table — a reusable, named, ordered row spec that renders consistently against the model's current state every time it's read (see get_table), instead of re-specifying a whole view from scratch on every call the way get_values_grid does. Each row is a dict with a row_type key selecting its shape: - {"row_type": "item", "name": ..., "label": null, "bold": false, "indent": 0, "negate": false, "value_format": null} — that line item's values across periods. label/value_format default to the item's own if left null. negate flips display sign only (e.g. show an expense as a positive "subtracted" number) — never changes the underlying computed value. - {"row_type": "label", "text": ..., "bold": true, "indent": 0} — section-header row, no values. - {"row_type": "blank"} — spacer row, no values. - {"row_type": "total", "label": ..., "line_item_names": [...], "bold": true, "indent": 0, "value_format": null} — sum of the named items per period, skipping a None ("not applicable") member rather than treating it as 0, same convention as a tag sum. - {"row_type": "tag_items", "tag": ..., "bold": false, "indent": 0, "include_total": false, "total_label": null} — expands to one item row per line item currently carrying tag (re-evaluated every render, so the table stays current as items are tagged/untagged with no need to edit it), optionally followed by a row summing them. There's no tag_total row type — if you want a tag's total as its own row without the individual members, give it a line item with a tag["x"][t] formula (see add_line_item) and reference that by name via a plain item row instead, the same way you would anywhere else a tag total is needed. - {"row_type": "percent_change", "name": ..., "label": null, "bold": false, "indent": 0} — period-over-period % change of an item, within the rendered period range (not necessarily the whole model). The first rendered period is always null (nothing to compare against); also null wherever the prior value is null/0 or the current value is null (undefined, not an error). - {"row_type": "cumulative_change", "name": ..., "label": null, "bold": false, "indent": 0, "as_percent": false} — change vs. the first rendered period (0, or 0% if as_percent, there). null if the base or current value is null, or the base is 0 and as_percent. Raises if another table already has this name, or if any row's name/line_item_names isn't a known line item yet — fix the name or add the item first. A tag_items row's tag is never validated to exist; an unused tag is fine, same as a tag["x"][t] formula tolerates one. A table may share its name with a scalar or line item (e.g. a debt_service_coverage table beside the line item of that name) — nothing references a table by name, so tables have their own namespace. Tools that take a bare name across kinds (set_notes, rename_item) then need kind="table" or kind="line_item" to say which one is meant.

clear_values

ChatGPT
Clear hardcoded values for a line item, handing those periods back to its formula (if any) — the counterpart to set_value's null, which sets a real, resolved "not applicable" value rather than un-setting a period. After clearing, a period with no formula behaves exactly like one that was never set — the call fails and rolls back cleanly (nothing changes) if that leaves something downstream unable to resolve, same as any other rejected mutation.

copy_model

ChatGPT
Duplicate a model into a new, independent one — for branching a scenario before trying something risky, without touching the original. Returns compact JSON: {"id": ..., "version": ..., "web_url": ...} for the new model (web_url links to it in the web dashboard). A literal copy of the source's current content, including its read_only flag and notes if set — a locked source produces a locked copy; call set_read_only(new_id, false) if you meant an editable branch instead. The copy goes in the same folder as the source. name defaults to "Copy of <source name>". You can only copy a model you own; the copy is yours too. The new copy's description is not carried over — it starts unset, since the source's own description would mislabel the copy with its parent's delta. set_description on the new model_id records what's different about this branch (e.g. "changed CPI assumption to 4%").

create_folder

ChatGPT
Create a folder for organizing models. Rejected if the name is blank, longer than 100 characters, or already used by another folder (ignoring case). move_model files models into it.

create_model

ChatGPT
Create and store a new, empty model. Returns compact JSON: {"id": ..., "version": ..., "web_url": ...}. frequency is fixed for the model's life: annual (the default — periods are years, [2027, 2028, ...]), quarterly or monthly. Give the periods as a list, or as start and end (inclusive): years for annual; labels for quarterly ("2027-Q1") and monthly ("2027-01"). Quarterly and monthly periods run consecutively, at most 80 quarters (20 years) or 120 months (10 years). In formulas t - 1 is always the previous period and t - 12 (monthly) / t - 4 (quarterly) the same period a year earlier — see add_line_item for the calendar helpers and how rates apply. id is what every other tool calls model_id — save it. The new model is owned by the identity this connection authenticated as. web_url links to the model in the web dashboard, where the user can see it as a spreadsheet with its tables and charts, and download it as an Excel workbook (also in get_model_summary). description is a short, stable one-liner for picking this model out of list_models later — what this model is (e.g. "Baseline water utility scenario, FY2026 budget, no rate increase"), not what's different about it relative to a sibling (that's what description means on a copy — see copy_model). For longer-form, evolving context about this model's current state or reasoning, use set_notes instead — description is for scanning a list, notes is for once you're already looking at it.

delete_folder

ChatGPT
Delete the folder called name (ignoring case). Refused, naming them, while any model is in it: move them elsewhere (move_model) or delete them first. Models in the trash don't count; restored, they come back in no folder.

delete_model

ChatGPT
Delete a stored model by moving it to the trash. It disappears from list_models, the web app and every other tool, and its view-only share link stops working. restore_model brings it back, exactly as it was, for 30 days; after that it's deleted permanently. list_models with trash=true lists the trash.

diff_models

ChatGPT
Structural and impact diff between two models. Returns what changed (formulas, hardcoded inputs, scalars, value/formula switches) and which shared line items' computed values differ as a result, flagged root vs propagated. Use get_values_grid to fetch the actual numbers for flagged items. Read-only; works on locked (read_only) models. Both models must have the same frequency (annual, quarterly or monthly). Compares only the periods both models have in common — see periods.only_a/only_b for what's outside that overlap. Returns a compact JSON string (not a dict — parse it); only non-empty sections/keys are included, so two identical models return just {"periods": {"both": [...]}}. Floats are rounded to 15 significant figures for display (never internally). Sections (all optional, omitted when empty): - periods: {"both", "only_a", "only_b"} — period lists. - items: {"only_a", "only_b"} — line item names present in one model but not the other. - scalars: {"only_a", "only_b", "changed": {name: [old, new]}}. - definitions (shared line items only): {name: {"formula": [old, new], "inputs": {period: [old, new]}, "switched": {period: ["formula"|"input", "formula"|"input"]}}} — formula when the formula text differs (after whitespace normalization); inputs when a period hardcoded in both models has a different value; switched when a period is hardcoded in one model but formula-driven in the other — reported even if the computed result happens to match (see impact). - impact (shared line items whose computed values differ, over the overlapping periods): {name: {"cause": "root"|"propagated", "first": period, "max_increase": {"value", "at"} | null, "max_decrease": {"value", "at"} | null}}. "root" means the item's own definition changed, or its formula directly references a scalar that changed; "propagated" means neither is true — its value only differs because something it depends on does. A switched period alone does not create an impact entry if the computed values still match. max_increase/max_decrease each report the largest signed b - a delta in that direction (max_increase.value is always positive, max_decrease.value always negative) and the period it occurs at; either is null if the value never moves that way. A differing period where either model's value is None contributes to first but has no numeric direction to report, so it never sets either. - metadata (shared line items/scalars): {name: [field names that differ, from "label"/"value_format"/"tags"/"notes"/"group"]} — field names only, never the actual text. - groups: {"only_a", "only_b", "label_changed"} — group names defined in one model but not the other, or whose label differs. Group order isn't compared (nor is line item order): it only affects layout.

export_json

ChatGPT
Export a stored model's content as a JSON string (for backup, or moving it elsewhere). Content only — read_only is a fact about this stored copy, not the model's own content, so a locked model exports the same as an unlocked one. Use list_models/get_model_summary to check lock status instead. Carries three envelope fields alongside the model's own content: format (always "proformaengine") and schema_version (the model JSON's shape version) — import_json uses them to recognize what it's being handed and to reject an export from a future, incompatible version of this server outright rather than failing confusingly partway through — and description, since that's storage-level metadata (like name), not part of the model's own content either, but still worth carrying along.

get_chart

ChatGPT
Render a stored chart against the model's current state, as compact JSON: {"title", "chart_type", "periods", "series", "left_axis_format", "right_axis_format"}. Each series: {"label", "source", "name", "tag", "type", "axis", "values", "value_format", "error"} — type is resolved (line/bar), values has one entry per period (null = not applicable), a tag series appears expanded into one entry per member, and a series naming something that no longer exists carries an error instead of failing the whole chart. periods overrides the chart's own range.

get_model_summary

ChatGPT
Return a snapshot of a model's structure — periods, scalars, line items — without their computed values by default, since those are usually the bulk of the payload and often not what you need just to see what a model contains. Pass include_values=True for the full picture, or use get_value/get_values/get_values_grid to pull specific results. Returns a compact JSON string (not a dict — parse it); floats are rounded to 15 significant figures for display (never internally). Includes web_url, a link to the model in the web dashboard, where the user can see it as a spreadsheet with its tables and charts. A line item whose formula is a pure tag sum (tag["x"][t], and nothing else — see add_line_item) carries a tag_members field listing exactly which line items it currently resolves to, so a miscategorized member (tagged the wrong thing, or not tagged at all) is visible right next to the formula rather than requiring a separate lookup. groups lists the model's groups (sections) in order, each with its label, notes, and members (line items then scalars). Line items are listed in section order — each group's items together, groups in order, ungrouped items last — and each line item and scalar carries its group (null when ungrouped). Without groups that's just the model's line item order. A line item's stale_input_periods lists any of its input_periods that aren't one of the model's current periods — normally empty. Non-empty either when values were staged ahead of adding a period (see set_value; they go live once set_periods adds it) or after set_periods narrows the model out from under a hardcoded value that was never removed (deliberately kept, not dropped — see set_periods). hardcoded_values/computed_values (with include_values=True) are always keyed in the model's own period order, with any such stale period's value appended after.

get_table

ChatGPT
Render a stored table against the model's current state — the payoff of add_table: the same saved row spec renders consistently every time, as the model changes, without re-specifying it from scratch the way get_values_grid requires. periods defaults to every period in the model; an explicitly-given period outside the model's own raises, same as get_values_grid. A row naming a line item that no longer exists (rare — remove_line_item already blocks removing one a table names directly; this can still happen via reset_model, which replaces the whole model without a per-item check, or a hand-edited import_json payload) reports an inline error on that row rather than failing the whole table. Returns a compact JSON string (not a dict — parse it): {"periods": [...], "title": ..., "rows": [{"row_type", "label", "bold", "indent", "values": [...] | null, "value_format", "error": str | null}, ...]} A row's value_format is a preset name, null, or — for a custom format — an object with decimals/thousands/prefix/suffix/ multiplier/scale/negative ("minus" | "parentheses")/zero ("number" | "dash" | "blank"). rows is an ordered list, not keyed by name (unlike get_values_grid's lines) — row order, and repeated/unnamed rows (label, blank, multiple totals), can't be represented as a name-keyed dict. values (when present) is aligned with periods. Floats are rounded to 15 significant figures for display (never internally).

get_value

ChatGPT
Get a scalar's value, or a line item's value at a period. The model is always freshly calculated (calculation is eager on every mutation). period is ignored for scalars, required for line items. Rounded to 15 significant figures for display (never internally), matching get_values/get_values_grid/get_model_summary.

get_values

ChatGPT
Get several values in one call — read-only, for pulling back a batch of results (e.g. after a set_values scenario) without one round-trip per value. One bad entry (unknown name, a line item queried without a period, a period the model hasn't computed) is reported per item rather than failing the whole call. Returns compact JSON: one result per input item, in the same order: {"name": ..., "period": ..., "value": float | None, "success": bool, "error": str | None}.

get_values_grid

ChatGPT
Get a table of values — several line items across several periods — in one call. More compact than get_values' per-(name, period) list when you want most or all periods for a handful of lines, since a name isn't repeated once per period. Selects line items the same way as get_model_summary: names for exact matches, tags for anything carrying one of them, combined as a union — at least one of the two is required. periods defaults to every period in the model; an explicitly-given period outside the model's own raises, rather than filling that cell with something ambiguous — null in a cell already means "not applicable" for that line at that period, so it can't also stand in for "this period doesn't exist." A name in names that isn't a known line item (including a scalar's name — this tool is for line items only) is reported in errors rather than failing the whole call; an unmatched tag is simply empty, not an error, same as get_model_summary. lines preserves order: names first, in the order given, then any further line item picked up only via tags (which has no inherent request order of its own), so a table can be built straight from the response without re-sorting. Returns a compact JSON string (not a dict — parse it): {"periods": [...], "lines": {name: {"label": ..., "value_format": ..., "values": [...]}}, "errors": [{"name": ..., "error": str}]} values is aligned with periods — same length, same order. Floats are rounded to 15 significant figures for display (never internally).

import_json

ChatGPT
Create a new stored model from a JSON string produced by export_json. Returns compact JSON: {"id": ..., "version": ..., "web_url": ...} (web_url links to the new model in the web dashboard). Always creates a new model (with a new model_id) rather than overwriting an existing one — use delete_model on the old one first if this is meant to replace it. An export missing format/schema_version (from before they existed) is treated as "proformaengine"/1 for backward compatibility. An export with a different format, or a schema_version newer than this server supports, is rejected outright with a clear error rather than failing confusingly partway through. A read_only key from an older export (back when the lock lived on the model itself) is honored as the new stored model's initial lock; current exports never contain one, since the lock lives at the storage layer now. A description key, if present, carries over to the new model the same way. For a monthly or quarterly model, periods and line items' values keys may be labels ("2027-01", "2027-Q1") instead of the stored period numbers an export contains; periods must fall in 1900-2200.

list_folders

ChatGPT
List this connection's folders, by name, as compact JSON: id, name, and model_count (models in it, not counting the trash). list_models with folder lists a folder's models.

list_models

ChatGPT
List models owned by this connection's identity (newest-updated first) as compact JSON: id, name, description, version, read_only, folder (its folder's name, or null), updated_at. folder lists only the models in the folder of that name (see list_folders). Models filed in an "Archive" folder are still listed; folders only organize, they never hide anything. trash=true lists the trash instead (see delete_model): id, name, description, deleted_at, and purge_on, when it's removed for good. A model in the trash can only be restored (restore_model).

move_model

ChatGPT
File a model in the folder called folder (ignoring case), or in no folder with null. The folder must already exist (create_folder). Doesn't change the model's content, version or last-updated time.

remove_chart

ChatGPT
Remove a chart. Nothing references a chart, so nothing blocks this.

remove_group

ChatGPT
Remove a group. Refused, naming its members, while any line item or scalar is still in it — move them with set_group (to another group, or group=null to ungroup) first.

remove_line_item

ChatGPT
Remove a line item from the model. Rejected up front, naming the dependents, if any other formula still references it by name, or if a table's item/total/percent_change/ cumulative_change row names it literally — a broken downstream table is exactly as real a dependent as a broken formula, not a silent surprise waiting in get_table. Tag membership alone doesn't block removal — neither a tag-sum formula's nor a table's tag_items row's — since both just lose a contributor rather than breaking, but if this item fed one, the response says so either way.

remove_scalar

ChatGPT
Remove a scalar from the model. Rejected up front, naming the dependents, if any formula still references it by name (get_model_summary shows every formula).

remove_table

ChatGPT
Remove a table. Nothing else in the model can reference a table — only the reverse (a table can reference a line item) — so there's no dependency check here, unlike removing a scalar/line item.

rename_folder

ChatGPT
Rename the folder called name (ignoring case). The models in it stay in it. Rejected if new_name is blank or another folder's name.

rename_item

ChatGPT
Rename a scalar, line item, table, chart, or group, and update every reference to it. One atomic change: every formula that references the old name (including an item's own self-reference like revenue[t-1]), every table row and every chart series naming it are updated to the new name, with each formula's text otherwise left exactly as it was. The item keeps its position, values, formula, tags, label and format. Computed values never change. Rejected, with the model untouched, if new_name is already taken or isn't a valid name (lowercase slug, not a reserved word). Scalars and line items share one set of names; tables, charts and groups each have their own, so a table may share a name with a line item. Renaming a group moves every member along with it. kind is only needed when old_name is such a shared name — the call is refused, asking for it, otherwise. Renaming one never affects the same-named other. The model keeps its id, unlike exporting, editing and re-importing it with import_json, which creates a new model. Not changed: tags (a category label, not a reference — tag sums keep working), labels, and free-text notes. If any note or label still mentions the old name, the response lists where. Afterwards only the new name is recognized.

rename_model

ChatGPT
Rename a stored model without touching its content or version.

reorder_groups

ChatGPT
Set the order of the model's groups — the order their sections appear in. names must list every group exactly once. Ungrouped items always come last. Purely cosmetic: never affects any computed value.

reorder_line_items

ChatGPT
Set the order of the model's line items — the order whole-model views (spreadsheet view, Excel export), get_model_summary, get_values_grid's tag-only matches, and tag sums' member lists all surface. Purely cosmetic: never affects any computed value. Without group, names must list every current line item exactly once — this reorders, it doesn't add, remove, or rename anything. With group, names must list exactly that group's line items, once each: only they move, within their own section — the usual way to fix the order inside one section of a grouped model. Order the sections themselves with reorder_groups, or let sort_by_dependency put a section into calculation order for you.

reset_model

ChatGPT
Replace a stored model's entire content with a fresh, empty model. Keeps the same model_id/name and frequency — only the content (scalars, line items, periods) is discarded. Give the new periods as a list or start/end, as in create_model. Raises if the model is currently read-only — covered by the same lock as everything else (set_read_only(False) unlocks it).

restore_model

ChatGPT
Take a model out of the trash (see delete_model), exactly as it was when deleted, view-only share link included. Only a model in the trash can be restored; list_models with trash=true lists them.

set_chart

ChatGPT
Replace an existing chart's whole definition (every argument, like add_chart — omitted ones reset to their defaults); get_chart returns the current definition, for changing only part of it. A chart is a named, reusable plot definition — like a table, it stores what to show, never values, and renders against the model's current state every time (get_chart; the web app draws it). Charts have their own namespace: a chart may share its name with a table or line item. series is an ordered list; each entry picks its source: - {"source": "item", "name": <line item>, "label": null, "type": null, "axis": "left"} — one line item's values. label defaults to the item's label. - {"source": "tag", "tag": <tag>, "type": null, "axis": "left"} — one series per line item currently carrying the tag (re-evaluated every render; an unused tag plots nothing). - {"source": "scalar", "name": <scalar>, "label": null, "type": null, "axis": "left"} — the scalar's value as a flat reference line (e.g. a covenant minimum); drawn as a line unless type says otherwise. chart_type ("line" or "bar") is every series' default; a series can set its own type to mix bars and lines. axis: "right" puts a series on a second y-axis, so dollars and ratios can share a chart. start_period/end_period (inclusive, optional; a year, or a label like "2027-03") limit the periods shown. value_format formats the left axis and right_axis_format the right; each defaults to the first series on that axis. Rejected, changing nothing, if an item/scalar series names something that doesn't exist, start_period is after end_period, or a format is an unknown preset. A chart that plots an item or scalar blocks removing it; renaming one updates the chart.

set_description

ChatGPT
Set (or clear, with null) a stored model's one-line description — shown in list_models, for telling apart models with similar names. See create_model's docstring for the description/notes distinction. Doesn't touch the model's content or version.

set_formula

ChatGPT
Update an existing line item's formula (pass null to clear it). See add_line_item's formula argument for the full grammar — arithmetic, comparisons, a ternary, min/max/abs/round, the ppmt/ipmt loan-amortization functions, and tag["x"][t] aggregation. keep_periods, if given, also clears every other hardcoded value on this item in the same call — e.g. converting a fully-hardcoded item to formula-driven with a single seed period is one call: set_formula(model_id, name, formula, keep_periods=[2019]), rather than a separate clear_values call after. Omit it (the default) to leave existing hardcoded values exactly as they are. See clear_values to clear values without also touching the formula. Leaves the line item's notes as they are, including any note explaining the old formula (set_notes changes them).

set_group

ChatGPT
File line items and/or scalars (any mix) under group, or pass group=null to leave them ungrouped — all in one atomic change, so organizing a whole model is one call per group. group must already exist (see add_groups). A line item moving into a group is placed at the end of that group's section; use reorder_line_items with group, or sort_by_dependency, to order the section. Display only: never affects any computed value.

set_label

ChatGPT
Set or clear the display label (title) of a scalar, line item, table, chart, or group. The label is what people see — headings and menus in the web app, a table's title, a group's section heading, Excel exports — while name stays the identifier used in formulas and every other tool (to change that, use rename_item). label=null (or blank) clears it, so the name is shown instead. kind is only needed when a table, chart or group shares name with something else — the call is refused, asking for it, otherwise. Never affects any value.

set_notes

ChatGPT
Set or clear free-text notes on the model itself, or on one scalar/line item/table/chart/group. notes is inert context for a reader (human or AI) — why an item exists, how it should be used, caveats — never read by calculation. name targets a scalar, line item, table, chart or group; omit it to set model-level notes (units, fiscal-year convention, sources, overall scope). notes=None clears whichever target this call points at. kind is only needed when a table, chart or group shares name with something else — the call is refused, asking for it, otherwise.

set_periods

ChatGPT
Replace the model's periods — as a list, or start and end (inclusive): years for an annual model, labels ("2027-01", "2027-Q1") for a monthly or quarterly one. An annual model's periods are processed in the order given — not validated or sorted; t is the year itself, so they should normally be consecutive years for t-1/t+1 to mean anything. A monthly or quarterly model's must run consecutively, in order, within 120 months / 80 quarters. Narrowing periods doesn't drop or reject any line item's now-out-of- range hardcoded values — they're kept as-is (e.g. so a temporary narrowing for editing doesn't destroy future-year assumptions entered earlier). The response says so when it happens; they also remain visible any time via get_model_summary's stale_input_periods. Growing periods back later makes them live again automatically. Widening periods (adding a year at either end) needs every line item with no formula to already have a value for each new period. Stage them first with set_values (it accepts periods outside the model), then call this; the response counts the staged values that went live. If any are still missing, the call fails, changes nothing, and names every missing item [periods].

set_read_only

ChatGPT
Lock or unlock a model against mutation. While read_only is true, every mutating tool on this model raises instead of changing anything — reads are unaffected. Always settable, including to unlock a model that's currently locked — otherwise there'd be no way back. This is a fact about the stored copy, not the model's own content: it doesn't change version, and it doesn't travel with export_json — an exported locked model is just data.

set_scalar

ChatGPT
Update an existing scalar's value. Leaves the scalar's notes as they are, including any note explaining the old value (set_notes changes them).

set_table_rows

ChatGPT
Replace an existing table's rows entirely — no partial row editing; resubmit the whole row list to change anything about it, including reordering. See add_table for the full row-type grammar. Same up-front validation as add_table: every item/total/ percent_change/cumulative_change row's line item reference must already exist.

set_tags

ChatGPT
Replace an existing line item's tags entirely (pass [] to clear them).

set_value

ChatGPT
Set (or overwrite) a hardcoded value for a line item at one period. value=null marks the period "not applicable" — a real, resolved value in its own right (resolves to null rather than erroring), distinct from the period never having been set at all. This is not a way to hand a period back to the item's formula — a null here still wins over the formula for that period, same as any other hardcoded value. See clear_values to actually clear a period and let the formula drive it. period is a year (annual) or a label ("2027-03", "2027-Q1"). It may be outside the model's periods: the value is staged — stored, but ignored until set_periods adds that period. That's how to add a year to a model with hardcoded inputs (an annual rollforward): stage every input's value for the new year here or via set_values, then call set_periods with the wider list. Leaves the line item's notes as they are, including any note explaining the old value (set_notes changes them).

set_value_format

ChatGPT
Set or clear how a scalar's or line item's values are displayed — in the web app, tables (unless a row overrides it), charts and Excel exports. value_format is a preset name ("usd", "usd_no_decimals", "accounting", "in_millions", "number", "number_two_decimals", "percent", "percent_one_decimal", "percent_two_decimals", "multiple", ...; an unknown name is rejected and the error lists them all) or a full ValueFormat; null goes back to the default number format. Display only: never affects any value. Table rows and charts have their own formats, set with set_table_rows/set_chart.

set_values

ChatGPT
Set several hardcoded values in one call — for loading/translating a model whose structure is already fully known, or running a scenario that overrides several assumptions at once, rather than one round-trip per value. Each update is applied independently through the same eager-calculate- and-rollback set_value: one bad entry (unknown name, a None that breaks downstream arithmetic, etc.) is reported as a per-item failure and doesn't block the rest. A value of null sets a real, resolved "not applicable" value — it does not hand the period back to the item's formula; see clear_values for that. A period outside the model's periods is staged until set_periods adds it, same as set_value — the usual first step of adding a year to a model with hardcoded inputs. Prefer set_value one at a time when building interactively — each call's own confirmation lets you catch a mistake immediately. Returns compact JSON: one result per input item, in the same order: {"name": ..., "period": ..., "success": bool, "error": str | None}, plus "staged": true on a successful entry whose period is outside the model's periods.

sort_by_dependency

ChatGPT
Reorder line items into calculation order — each after the items it reads in the same period, so inputs come first and a section reads top to bottom like a well-built spreadsheet. Works within each section of a grouped model (sections and group membership never change), or only within group if given; an ungrouped model is one section. Stable: only out-of-order items move, so a section already in order is untouched and a manual arrangement survives as far as dependencies allow. Earlier-period references (x[t-1]) don't count, so an opening/closing balance pair sorts cleanly. A one-off rewrite of the stored order, not a standing mode. Never affects any computed value. Returns which items moved.

add_chart

Claude

add_groups

Claude

add_line_item

Claude

add_line_items

Claude

add_scalar

Claude

add_scalars

Claude

add_table

Claude

clear_values

Claude

copy_model

Claude

create_model

Claude

delete_model

Claude

diff_models

Claude

export_json

Claude

get_chart

Claude

get_model_summary

Claude

get_table

Claude

get_value

Claude

get_values

Claude

get_values_grid

Claude

import_json

Claude

list_models

Claude

remove_chart

Claude

remove_group

Claude

remove_line_item

Claude

remove_scalar

Claude

remove_table

Claude

rename_item

Claude

rename_model

Claude

reorder_groups

Claude

reorder_line_items

Claude

reset_model

Claude

set_archived

Claude

set_chart

Claude

set_description

Claude

set_formula

Claude

set_group

Claude

set_label

Claude

set_notes

Claude

set_periods

Claude

set_read_only

Claude

set_scalar

Claude

set_table_rows

Claude

set_tags

Claude

set_value

Claude

set_value_format

Claude

set_values

Claude

sort_by_dependency

Claude

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/proformaengine"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/proformaengine)