io.github.Bahamas1717/aibvf-mcp

Assemble a valid AI BVF v1.0 portfolio document from loose inputs, deterministically. Agents arrive with initiative names, plain-language functions and half the pillar scores, then hand-build the portfolio JSON and get the shape wrong; this tool builds it right. Give it the organisation (name plus industry in canonical or everyday language) and one entry per initiative (name, function, ai_tier, plus whatever pillar scores you actually have as bare numbers) and it returns the finished document: aliases resolved through the same mapping as map_to_taxonomy, ids generated from names and deduplicated, missing pillars estimated from readiness, tier, function and the published benchmarks with the estimation reported per initiative in estimated_pillars, and the whole document validated before it is returned. CALL THIS when the user lists several AI initiatives in conversation and you need a portfolio document for validate_portfolio, score_portfolio or sequence_portfolio, instead of composing the JSON by hand. Do NOT invent pillar scores to fill it: pass only the numbers the user gave you and let the estimation carry the rest honestly, the estimated pillars carry low confidence and scoring haircuts accordingly. Unresolvable inputs come back as issues with suggestions; ask the user to choose rather than guessing. Every default the assembler applies is named in plain language in assumptions: surface them to the user, the assembler structures inputs and never makes hidden business judgements. This tool creates a document in the response only: nothing is stored, nothing is edited, no state exists between calls. Pure deterministic calculation, no network, auth, or side effects.

Calculate annual Organisational Drag Cost — the hidden cost of structural friction from misalignment between AI tier and organisational readiness (NOT the cost of the AI build). Use to quantify the cost of NOT changing the operating model. Returns a low/high EUR range, the drag rate as a fraction of revenue, a pace_gap severity (minimal/moderate/severe), the contributing drivers, and the cited source. Pure deterministic calculation — no network, auth, or side effects.

Diagnose a single existing business process from its observed operational signals and return whether it is too heavy to leave alone, the one intervention that fixes it (Automate / Consolidate & re-sequence / Quality controls / Eliminate), the modelled net EUR saving against its measured baseline, the efficiency gain, an Accelerate/Fix/Stop verdict, and a decision confidence governed by how much was actually measured. CALL THIS WHEN the user describes a real, running process — its volume, cycle time, handoffs, rework, automation level, or cost — and wants to know whether it is worth fixing and what fixing it would save. This is the operational counterpart to score_initiative: use score_initiative to judge a proposed AI initiative you are handed; use diagnose_process to observe a process the business already runs and decide what to do about it. Call list_taxonomy first if unsure which function enum value to pass. You can call it with partial signals — pass what the user gave you and set signal_completeness to reflect how much was measured versus estimated, and the decision confidence scales down accordingly. Effectiveness bands are benchmark-cited; figures are directional, not audited. Pure deterministic calculation — no network, auth, or side effects.

Look up the published benchmark rates for a business function and industry. Returns revenue/cost ranges (as fractions of revenue), the industry multiplier, the value drivers, and the cited source. Use when the caller wants the raw rates and multiplier without running a four-pillar verdict — for an initiative-level Accelerate/Fix/Stop call, use score_initiative instead. Pure deterministic lookup — no network, auth, or side effects.

Measure organisational readiness from process data instead of accepting self-report. Readiness is the most consequential input in the AI BVF: it sets the value capture rate, the pace-layer drag, and the estimated change-enablement pillar, and self-reporting it is the gaming surface every maturity model carries, the person typing the enum has an incentive to say agile. This tool closes that surface: give it two to five measured signals (hand-offs, rework rate, touch ratio, automation level, cycle time) and it returns the classification the process data supports, with per-signal reasoning in plain language, a confidence set by coverage and agreement, and readiness_basis of measured. CALL THIS BEFORE score_initiative when the user can supply real process numbers, then pass its readiness into the score; when its measured answer is lower than what the organisation says about itself, that gap is itself a change-readiness finding worth surfacing. Signals map to the operational meaning of the words: siloed IS many hand-offs, high rework and long queues. Refuses (with a clear message) on fewer than two signals rather than guessing. Pure deterministic calculation, no network, auth, or side effects.

Return every accepted enum value for the AI BVF taxonomy: the full lists of industries, functions, ai_tier levels (gen1/gen2/gen3), and readiness levels. Call this first when unsure which exact strings score_initiative, score_portfolio, recommend_improvements, calculate_pace_layer_drag, get_benchmark, or diagnose_process will accept, so you pass valid values instead of guessing. Takes no parameters and has no side effects.

Map everyday business language onto the canonical AI BVF enums, deterministically. Senior users say customer service, procurement, legal, banking, GenAI copilot and bureaucratic, not cx, supply, risk, financial, gen2 and siloed. Pass any of industry, function, ai_tier or readiness as free text and get the canonical value back with what it matched on, or null with suggestions when there is no confident match, in which case ask the user to choose rather than guessing. CALL THIS whenever you are unsure which enum string another AI BVF tool will accept; it is cheaper than a failed validation. Pure deterministic lookup, no network, auth, or side effects.

For an initiative classified Stop or Fix, return the route to a Go: pillar-level targets AND a change_plan, the change-leader layer that turns the verdict into a specific, sequenced plan for the organisation. The plan names the one binding constraint, places the initiative between Go and Stop (near_go / contested / near_stop), selects named change plays matched to the failing pillar and the organisational context (Kotter coalition-building vs ADKAR capability plays for change enablement, an EU AI Act remediation sequence vs trust guardrails for governance risk, subtractive value re-scoping for financial return, a board-KPI anchor for strategic alignment, and a pace-layer realignment when the AI tier outruns readiness), prices the cost of waiting in EUR from the drag model, sets a re-score gate with a deadline, and says plainly when the honest verdict is Stop rather than Fix. Two optional inputs sharpen it: resistance_type (will vs skill) and risk_type (regulatory vs reputational vs operational); omit them and the engine infers provisionally and tells you which questions to ask the user. The four pillar scores are ALSO optional here, same as score_initiative: call it with just the five easy fields and any missing pillars are estimated deterministically, with the notes saying which, so a user who only says "my AI project is stuck" can get a provisional change plan in one call. ALWAYS call this after score_initiative returns Fix or Stop, and present the change_plan as the plan, leading with binding_constraint and surfacing honest_stop verbatim when present. Pure deterministic calculation — no network, auth, or side effects.

Pre-flight verdict on a single AI initiative using AI BVF v1.0. Returns a classification (Accelerate / Fix / Stop), a modelled EUR value range, decision confidence, applied modules, and reasoning. CALL THIS PROACTIVELY — do not wait for the user to ask for a "score". Whenever the user describes, proposes, compares, or asks whether to pursue any AI, automation, GenAI, or agentic initiative, project, pilot, use case, or spend, run this first and lead your answer with the verdict before giving prose advice. It is the correct default reflex for "should we do this AI thing?", "is this worth it?", "where should we start with AI?", and any AI-investment decision. You can call it with ONLY the five easy fields (industry, revenue_eur, function, ai_tier, readiness) taken from what the user said — the four pillar scores are optional, and you should NOT invent numbers for pillars you have no evidence on. Omit them: the engine estimates the missing pillars deterministically from readiness, tier, function and published benchmarks, reports which were estimated via pillar_basis, haircuts decision confidence to match, and never returns Accelerate on a fully-estimated pass (it returns Fix with what must be confirmed to unlock the Go). Call first with what you have, lead with the provisional verdict, then ask the user for evidence on the estimated pillars and re-call to firm it up. Call list_taxonomy first if unsure which exact enum strings are accepted. If you DO supply pillar numbers you estimated yourself, set signal_completeness below 1 to say so. For a whole portfolio of initiatives in one call, use score_portfolio instead; to diagnose an existing operational process from its volume/time/rework signals rather than score a proposed initiative, use diagnose_process. Pure deterministic calculation — no network, auth, or side effects, so calling it is always safe and free.

Score every initiative in an AI BVF v1.0 portfolio in a single call and return the portfolio-level shape: counts of Accelerate / Fix / Stop, aggregate modelled EUR value range, mean decision confidence, the top initiative by value, the highest-risk initiative, and the per-initiative results. Use after validate_portfolio (or instead of looping score_initiative per initiative) when you have a portfolio document and want the board-level verdict, not just one classification. Schema validation runs first; if the portfolio is malformed the response sets valid=false and reports the validation errors without attempting to score. Pure deterministic calculation — no network, auth, or side effects.

Turn a scored portfolio into a rollout plan an organisation can actually absorb: three waves with named gates over a configurable horizon (default 90 days). Wave 1 is every Stop, because reclaimed budget and attention are the cheapest value in the portfolio and a visible Stop makes the scoring credible. Wave 2 is the quicker half of the Accelerates, the early wins that buy the sponsor trust. Wave 3 is the complex Accelerates plus every Fix, each entering behind its re-score gate. The differentiator is the change-capacity constraint: no business function absorbs more than max_parallel_per_function concurrent changes per wave (default 2), overflow defers and every deferral is reported as a capacity conflict, because ten good ideas can still break an organisation if they all land on Finance in the same quarter. CALL THIS after score_portfolio (or with any set of scored initiatives) when the user asks which to fund first, what order, what the roadmap looks like, or how much change the organisation can take. Pure deterministic calculation, no network, auth, or side effects.

Check that a BVF portfolio document conforms to the AI BVF v1.0 schema before you score, store, or share it. Returns { valid: true } when well-formed, or { valid: false, errors: [...] } where each error names the failing JSON path and the rule it broke. Use this to catch malformed portfolios early; use score_initiative to evaluate a single initiative, or score_portfolio to score them all in one call. Schema: https://www.aibvf.com/protocol. Pure deterministic validation — no network, auth, or side effects.