Populate a request list from available informationA populated workbook, each answer sourced, the blanks left as the ask.2 steps
The same list answered from what you already hold, a data room export, filings, or an earlier session, so the ask left to the target is only what is genuinely missing.
Turns a target profile into a prescriptive diligence agenda: the topics to cover, the questions to ask under each, and the attention areas the profile makes material.
enum. modern-cloud-native · hybrid-legacy · self-managed-infra · datacenter-vendor. The largest single driver of which questions surface.
headcount
Example
ordinal enum. Company size band. Questions can gate on a minimum band rather than an exact match.
revenueRange
Example
ordinal enum. Revenue band, gated the same way.
growthStage
Example
enum. Where the company sits in its funding and growth arc.
companyAge
Example
ordinal enum. Age band, gated the same way.
geographies
Example
enum array. The only multi-select input. Two or more specific regions auto-adds multi-region.
businessModel
Example
enum. How the target earns.
scaleIntensity
Example
enum. Load and growth pressure on the platform.
transformationState
Example
enum. How much change the estate is already carrying.
dataSensitivity
Example
enum. Drives the privacy and security question set.
operatingModel
Example
enum. How engineering and operations are organized.
All thirteen dimensions are required: the engine has no defaults, and a missing field is rejected before it runs. Each accepts a fixed identifier set; call the tool with an invalid value and the error names the valid ones.
Searches GST's anonymized engagement record by free text, theme, and engagement side, so a current deal can be framed against work already done.
Arguments
click a value to copy it
Name
Example
Requirement
search
Example
string, optional. Free text matched against code name, industry, summary, and technologies. Substring match, not fuzzy.
theme
Example
string, optional. One of the values list_portfolio_facets returns, or "all". Defaults to "all".
engagement
Example
string, optional. Buy-Side · Sell-Side · all. Defaults to "all".
The empty call returns every engagement. There is no limit argument: the tool mirrors the website filter surface exactly, and the website renders the full set.
Queries 123 regulatory frameworks by jurisdiction, category, and free text. The curated corpus carries current effective dates, scope language, key requirements, and statutory penalty bands.
Arguments
click a value to copy it
Name
Example
Requirement
jurisdiction
Example
string or string array, optional. Lowercase codes: "eu", "us", "us-ca", "ca-qc". An array combines matches in one call.
string, optional. Matches name, curated aliases, summary, and id. Common short forms resolve to the statute they name.
limit
Example
number, optional. Default 20, maximum 120.
Keep limit at or near its default and narrow by category. Broad multi-jurisdiction queries return very large responses, and raising the limit is how a result outgrows a client rather than how it gets more useful. When an array holds more than one value, the response omits that filter from its deeplink, because the website uses single-select chips.
Lists the jurisdictions and categories the regulatory corpus actually indexes, plus the total framework count. The recovery call when a jurisdiction code does not resolve.
Arguments
Takes no arguments.
Example
complete and valid as written
list_regulation_facets({})
Returns
jurisdictions and categories. Every value search_regulations will accept.
totalFrameworks. The size of the indexed corpus.
Availability
Included in all tiers. Rate ceilings are per-client capability limits, not contractual quotas.
Scores cloud cost governance maturity across six domains and returns a prioritized improvement list. Wraps the engine behind the Hub wizard.
Arguments
click a value to copy it
Name
Example
Requirement
answers
Example
map of question id to a score. Keys take the form q1_1, q1_2 … q6_N across the six domains.
companyStage
Example
enum, optional. Canonical funding stage (seed · series-a · series-b · series-c · pe · enterprise) or one of the four native bands. Adds a benchmark band; never changes the score.
The scale is 0 Not in place · 1 Ad hoc · 2 Established · 3 Optimized, with -1 meaning Not sure. The map is sparse: missing questions count as zero, so a partial assessment reports an honest absence of information rather than failing. Only an explicit -1 is reported as skipped.
overallScore and maturityLevel. Reactive · Aware · Optimizing · Strategic.
domainScores. One per domain, with the raw and normalized score.
showFoundationalFlag. Set when visibility or attribution scores low, independent of the overall score.
recommendations. Sorted by impact, then effort, then domain.
answeredCount, totalQuestions, skippedCount. How much of the assessment was actually answered.
deeplink. Opens the wizard on the results view with the same answers.
A high overall score can hide a foundational gapTwo of the six domains, visibility and tagging plus account structure and attribution, are foundational: everything else rests on them. If either scores at or below a third, the result raises its foundational flag no matter how strong the overall number looks.
Availability
Included in all tiers. Rate ceilings are per-client capability limits, not contractual quotas.
Benchmarks total technology cost against stage-adjusted ranges and projects a 36-month trajectory. Wraps the TechPar engine behind the Hub wizard.
Arguments
click a value to copy it
Name
Example
Requirement
arr
Example
number > 0. Annual recurring revenue, dollars. Drives every percentage-of-revenue calculation.
stage
Example
enum. Canonical funding stage: seed · series-a · series-b · series-c · pe · enterprise. Selects the per-stage benchmark band.
mode
Example
enum, no default. quick reads rdOpEx directly; deepdive synthesizes R&D OpEx as engCost + prodCost + toolingCost.
capexView
Example
enum. cash includes rdCapEx in totals; gaap excludes it.
growthRate
Example
number. Annual revenue growth, as a percentage; drives the 36-month projection.
exitMultiple
Example
number, zero or more. Translates the cumulative gap to exit value. 12x is the SaaS convention.
infraHostingAnnual
Example
number, zero or more. Annual infrastructure and hosting cost, dollars. Must be above zero.
infraPersonnel
Example
number, zero or more. Annual infrastructure personnel cost, dollars.
rdOpEx
Example
number, zero or more. R&D OpEx. Read in quick mode; discarded in deepdive.
rdCapEx
Example
number, zero or more. Capitalized R&D.
engFTE
Example
number, zero or more. Engineering headcount; yields revenue per engineer.
engCost
Example
number, zero or more. Annual engineering personnel cost. deepdive only.
prodCost
Example
number, zero or more. Annual product personnel cost. deepdive only.
toolingCost
Example
number, zero or more. Annual tooling cost. deepdive only.
All 14 fields are required in both modes; the engine ignores the fields the selected mode does not read. All money fields are annual dollars on one currency basis. A zero arr or infraHostingAnnual returns invalid-input, never a stack trace.
Estimates the annual carrying cost of accumulated technical debt from team size, salary, maintenance burden, and delivery cadence.
Arguments
click a value to copy it
Name
Example
Requirement
teamSize
Example
integer above zero. Engineering headcount. A direct multiplier on the carrying cost.
salary
Example
number above zero. Average fully loaded annual engineering salary, dollars.
maintenanceBurdenPct
Example
number 0 to 100. Share of engineering capacity consumed by maintenance and debt servicing. The headline input.
deployFrequency
Example
enum. Multiple/day · Daily · Weekly · Bi-weekly · Three-week · Monthly · Quarterly+ · Bi-annually · Annually. Sets the DORA tier and the velocity multiplier.
incidents
Example
integer, zero or more. Production incidents per month.
mttrHours
Example
number, zero or more. Mean time to recovery, hours per incident.
remediationBudget
Example
number, zero or more. Capital available for debt paydown, dollars.
arr
Example
number, zero or more. Annual recurring revenue, used to express the cost as a share of revenue.
remediationPct
Example
number 0 to 100. Expected reduction in debt cost from the remediation.
contextSwitchOn
Example
boolean. Whether to model the context-switching overhead surcharge.
These are business values, not wizard slider positions: pass the precision you actually have. An arr of zero is allowed and reports the revenue share as zero rather than dividing by it.
Builds the blank information request list as an XLSX workbook, configured for the engagement. This is the ask GST hands a target before diligence tools can run.
Arguments
click a value to copy it
Name
Example
Requirement
targetName
Example
string, optional. The target the list is addressed to.
companyName
Example
string, optional. Composed into the workbook title.
projectName
Example
string, optional. Composed into the title alongside companyName.
transactionContext
Example
enum, optional. sell-side · buy-side · value-creation · unknown. Fires the authored skip-if directives, removing questions that do not apply.
includeSections
Example
array of two-digit section ids, optional. Keeps whole sections. Defaults to all ten.
excludeRequests
Example
array of NN-II keys, optional. Removes individual questions. Call list_irl_requests to find the keys.
customRequests
Example
array, optional. Extra questions, each with a section and its text.
showCanonicalReference
Example
boolean, optional. Whether to print the canonical reference column.
Every field is optional: the empty call produces the full canonical workbook. The three subtractions compose, and surviving questions keep their reference ids, so the gaps read as deliberate rather than as an incomplete list.
Builds the same workbook already populated from evidence the model holds, writing each answer and the source it rests on into the row. Removes the wait for a third party wherever the information already exists.
Arguments
click a value to copy it
Name
Example
Requirement
fills
Example
array of 1 to 200 entries, each { ref, fileLocation, comments }. All three are required per entry.
ref
Example
The workbook Reference value for the row being answered, for example 0-03. Must exist in the configured workbook, and cannot repeat.
fileLocation
Example
What the answer rests on: a document reference and optional locator, or a bracketed origin note. A reference, never an excerpt.
comments
Example
The answer itself, as single-line prose. Under the extraction rules an entry here is a real answer, never a placeholder.
targetName
Example
string, optional. Same scoping arguments as the blank generator, minus productSummary.
Requiring both a source and an answer on every entry is structural: a row cannot be answered without saying what it rests on. The shape of the reference is checked; whether the referenced document exists is deliberately not, because that is what the human reviewer verifies.
filename and the workbook bytes, identical in behavior to a target-returned file.
filledRowCount and filledRefs. The operator review checklist.
blankRowCount. What remains to ask for, which is the follow-up list.
Blank rows are the askRows the evidence cannot support stay empty on purpose. Filling them from inference would produce a workbook that looks complete and is not, so the unanswered rows are what goes back to the target.
Availability
Included in all tiers. Rate ceilings are per-client capability limits, not contractual quotas.
Returns the canonical question set behind the information request list, with the key for each question. The only way to map "drop that question" onto the key the generator accepts.
Arguments
Takes no arguments.
Example
complete and valid as written
list_irl_requests({})
Returns
requests. Each with its NN-II key, section, section title, question text, and any engagement contexts that auto-remove it.
sectionCount and bulletCount.
Availability
Included in all tiers. Rate ceilings are per-client capability limits, not contractual quotas.
Registers a completed information request list with the server and returns a short hash for it. Everything downstream in the dossier pipeline refers to the document by that hash instead of resending it.
Arguments
Name
Requirement
filledIrl
string, required. The populated list, entire markdown body, at least 200 characters.
The hash is computed byte for byte with no normalization, so the same body always produces the same hash. Do not hand-compute it: the downstream tools accept only the value this call returns.
Checks every claim in a draft back to the request list it cites, and reports which citations actually hold. Run it to see the verdicts before they are written into a deliverable.
Arguments
Name
Requirement
citations
array, required. Each entry pairs the claim path with the citation supporting it. A citation may be one string or up to eight.
irlBodyHash
string. The hash from prepare_irl_body.
filledIrl
string. The body itself, as an alternative to the hash.
One of irlBodyHash or filledIrl must be supplied; the body wins when both are. Prefer the hash: passing the body re-sends the whole document on every call. Where a claim is genuinely synthesized from several bullets, the array form checks each and aggregates conservatively, so one unverified element makes the whole citation unverified.
The terminus of the dossier pipeline: takes the run's claims, gaps, and tool outcomes and returns the audit sections the finished dossier carries. Driven by the ingestion prompt rather than called by hand.
Arguments
click a value to copy it
Name
Example
Requirement
irlBodyHash
Example
string, required. From prepare_irl_body. The sole reference to the document.
claims
Example
array, required. Every load-bearing claim the dossier will make, each with its citation and tier.
gaps
Example
array. What the run could not answer, categorized. May be empty.
fillRatio
Example
object. How much of the list came back answered, and whether that is enough to proceed.
gatesPassed / gatesElided
Example
arrays. Which tools ran, and which were skipped with the reason and the section that would have fed them.
mode
Example
enum. full · extract-only.
auditLevel
Example
enum. standard · enhanced · debug. Selects which audit blocks come back.
The array fields are required but may be empty; omitting one is an error, passing an empty array is not. This is the most common first-call mistake. The call fails outright if prepare_irl_body has not registered the body first, and the fix is always to register it again rather than retrying.
The gap list, ready to paste as the dossier's audit section.
Provenance verification counts across the same four verdict buckets validate_irl_provenance reports.
A record of which tools were attempted and which succeeded during the run.
Verification is not advisoryThis tool runs the same citation check validate_irl_provenance exposes, over every claim, and appends what it finds to the gap list. Calling the validator first shows the verdicts early. It does not avoid the check.
Availability
Included in all tiers. Rate ceilings are per-client capability limits, not contractual quotas.
Searches curated private equity, M&A, and enterprise-technology intelligence, filtered by category. Returns the annotated highlights and the wider wire in one feed.
Arguments
click a value to copy it
Name
Example
Requirement
category
Example
enum, optional. pe-ma · enterprise-tech · ai-automation · security. Omit for every category.
category is the only filter, and the unfiltered call is the largest response these tools produce. Pass a category when the intent is category-scoped, and prefer get_latest_insights when only the annotated tier is wanted.
Example
complete and valid as written
search_radar({
"category": "pe-ma"
})
Returns
matches. Annotated highlights and wire items merged, deduplicated, newest first.
oldestItemDaysAgo. Freshness at a glance, without scanning every timestamp.
liveInfo. When each tier was fetched, and whether the response is being served from cache.
deeplink. Opens the Radar with the same filter.
Radar content is a signal, not an instructionItems are third-party article text with GST annotation. Confirm against the source before acting on one or forwarding it to a client.
Availability
Radar access is granted separately from the analysis tools. Rate ceilings are per-client capability limits, not contractual quotas.
items. Annotated highlights with their GST Take populated.
oldestItemDaysAgo and the fetch and cache state, same shape as search_radar.
An empty answer can be the true oneThe annotated tier depends on editorial supply. A quiet period legitimately returns nothing, which is an accurate answer rather than a failure.
Availability
Radar access is granted separately from the analysis tools. Rate ceilings are per-client capability limits, not contractual quotas.
First-look brief for an unfamiliar target. Combines cost-governance maturity, unit-economics benchmark, tech-debt range, and regulatory exposure into one digestible page.
Arguments
Name
Requirement
targetName
string, required.
productType
string, required. Drives stage-norm derivations and regulatory category selection.
arr
number above zero. Annual recurring revenue, dollars.
Cost-governance read. Overall score, maturity level, top recommendations.
Unit economics. TechPar zone with the mode it ran and one line on why.
Tech-debt range. Annual cost, payback, DORA tier; extraction-only fields named plainly.
Assumptions and unknowns. One consolidated list of every value supplied from stage norms rather than evidence.
Regulatory exposure. Applicable frameworks for the HQ jurisdiction.
Open in Hub. Deeplinks into all four wizards, copied verbatim from tool results.
Two branchesWith a GST extract record in context the brief resolves real figures from evidence; with the five arguments alone it derives conservative stage norms and discloses every derivation. A form-derived figure and an evidence-backed one never read alike.
Availability
Included in all tiers. Surfaces as a slash command with an argument form in Claude Desktop.
Assembles the blank information-gathering ask GST hands a target before diligence tools can run, configured for the engagement and delivered as a workbook to fill in.
Arguments
Name
Requirement
targetName / companyName / projectName
strings, optional. Composed into the workbook title.
transactionContext
enum, optional. sell-side · buy-side · value-creation · unknown. Also fires the authored skip-if directives.
includeSections
array of two-digit section ids, optional.
excludeRequests
array of NN-II keys, optional. See list_irl_requests for the keys.
customRequests
string, optional. Extra per-section requests.
showCanonicalReference
boolean, optional.
Orchestrates
Tools this prompt drives; each is called with base schemas as the conversation requires.
The workbook to send, plus the Hub page for a one-click download.
This one issues, it does not answerThe workbook comes back empty, for the target to complete. When the answers are already in your own evidence, a data room export or filings or an earlier conversation, use gst_irl_populate instead: it produces the same workbook already populated, and leaves only the rows it could not support blank.
Availability
Included in all tiers. Surfaces as a slash command with an argument form in Claude Desktop.
Populates the request list from evidence already in the conversation, a data room export, filings, or prior sessions, instead of waiting for the target to return one. Stops at the artifact for the operator to review.
Arguments
Name
Requirement
targetName / companyName / projectName
strings, optional. Composed into the workbook title.
An inventory of the evidence found, before anything is written.
The populated workbook, each answered row carrying the source it rests on.
The blank rows, which are the remaining ask to the target.
It stops at the artifactThe prompt does not run the diligence sweep. The operator reviews the populated workbook first, then runs gst_irl_sweep on it exactly as for a target-returned one. Re-running with new evidence extends the file rather than overwriting it.
Availability
Included in all tiers. Surfaces as a slash command with an argument form in Claude Desktop.
Distills a populated request list into a portable extract record plus the per-tool payloads derived from it. Makes no tool calls, so the record can be saved and pasted into later sessions and other GST prompts.
Arguments
Name
Requirement
filledIrl
string, optional. The populated list, entire markdown body. Omit it when the list is attached to the conversation or was pasted earlier.
Pasting into a single-line client field collapses the line breaks. The run still works: what the record is checked against normalizes whitespace before matching.
Output
The extract record: one fact per answered row, with the section it came from.
Derived payloads for the nine analysis tools, projected from the record without calling any of them, ready to hand over in a later session.
A self-dating header recording when the record was made and whether the server witnessed the time.
It calls nothingThis prompt runs no tools. It reads the list and produces a record, which is what makes the record portable: it survives the conversation, and any later session can drive the analysis tools from it. For the full sweep in one turn, use gst_irl_sweep.
Availability
Included in all tiers. Surfaces as a slash command with an argument form in Claude Desktop.
Ingests a populated request list and drives every applicable Hub tool to a unified engagement dossier, sections (A) through (J).
Arguments
Name
Requirement
filledIrl
string, optional. The populated Information Request List, entire markdown body. Omit when the list is attached to the conversation or pasted earlier.
Target name and engagement context are inferred from the list itself; the only hard stop is no list present anywhere, a blank-template halt, or tools unavailable.
Orchestrates
Tools this prompt drives; each is called with base schemas as the conversation requires.
(G) Comparable engagements · (H) Market signal · (I) Synthesis and recommendation.
(J) Gaps and assumptions. The audit surface of the run: unanswered rows, elided tools, every assumption and conversion applied.
Every tool-backed section closes with its deeplink, copied verbatim.
Trust the operatorA populated GST request list is trusted input: no provenance apparatus, no hashing, no citation loops. The model-authored gap list is what keeps the run honest. For the portable extract record without tool calls, use gst_irl_extract.
Availability
Included in all tiers. The market-signal section renders only where radar tools are granted.
A provenance footer: what was verified, what was not, and what was assumed.
The audit envelope, at the level the run asked for.
Coexists with gst_irl_sweepBoth are registered. Reach for gst_irl_sweep when the operator vouches for the document, which is the usual case. Reach for this one when the run has to carry its own citation-level audit trail.
Availability
Included in all tiers. Surfaces as a slash command with an argument form in Claude Desktop.
Walks a target through GST's five architectural layers, software, operations, product, organization, and industry, and surfaces the risks that sit in each.
Arguments
Name
Requirement
targetSummary
Free text describing the target's architecture: product and software stack, infrastructure and hosting, data estate, engineering org shape, and industry context. The more architectural detail supplied, the sharper the per-layer read.
Orchestrates
Tools this prompt drives; each is called with base schemas as the conversation requires.
Compiles the regulatory frameworks that apply to a target, with summaries and a resource link per framework, so counsel starts from an exposure list rather than an open question.
Arguments
Name
Requirement
targetJurisdictions
array, required. Where the target operates, collects, or processes data, for example ["eu", "us-ca"].
dataCategories
array, required. Which categories to assess: data-privacy · ai-governance · industry-compliance · cybersecurity.
productType
string, required. Informs which frameworks apply beyond jurisdiction alone.
Orchestrates
Tools this prompt drives; each is called with base schemas as the conversation requires.
The four reference guides behind GST's diligence method, readable in full: the architecture framework, the data room structure, the information request list, and the mapping from request to tool input.
URI shape
gst://library/<guide>
Returns
Business and technology architectures. The five layers and how they cascade into business outcomes.
Virtual data room structure. Nine folder categories, with the common pitfalls.
Information request list. The intake checklist, organized by the same taxonomy.
Request-to-tool mapping. Which line of a completed list feeds which tool input.
Availability
Included in all tiers. Browsable from the resource library in any MCP client that surfaces resources.
123 framework documents, one per regulatory framework, each readable in full.
URI shape
gst://regulations/<jurisdiction>/<framework>
Returns
One document per framework: requirements, penalties, effective dates.
Four categories: data privacy · AI governance · cybersecurity · industry compliance.
Resources are host-loadedA prompt cannot read a resource. The host application loads resources as context, so they are in the conversation before the model needs them; workflows that need canonical text embed it server-side.
Availability
Included in all tiers. Browsable from the resource library in any MCP client that surfaces resources.
OAuth 2.1 for both shapes of client: client credentials for a machine client, and authorization code with PKCE for a person connecting a desktop AI client.
Returns
Machine clients receive a client id and secret, and exchange them for a short-lived access token.
A person connecting a desktop client is taken through a consent screen and never handles a token by hand.
Every call carries its token; there is no anonymous surface beyond the health check.
Clients are registered, not self-serviceThere is no dynamic client registration and no user directory. Credentials are issued through a conversation with GST, which is also how scopes and a tier get set correctly the first time.
Availability
Request access to be provisioned. See the request-access path on the MCP Server page.
Per-client capability ceilings by tier, applied per minute and per day, with the radar tools metered separately from the analysis tools.
Returns
Every response carries standard rate-limit headers, including the caller's own ceilings, so a client can self-diagnose its budget without first hitting a limit.
A warning surfaces as the budget approaches its ceiling, ahead of any refusal.
Over the ceiling, calls are refused with the time to wait rather than failing opaquely.
Tunable, not contractualThese are abuse and capacity limits, adjustable per client, not ratified service quotas. No rate commitment is contractually made. The current per-tier ceilings are published on the MCP Server page.
Availability
Applies to every tier. A tier change takes effect on the next window.