The pre-flight check for agentic AI

Stop bad AI projects before
agents recommend them.

The scoring tool your Claude agent calls before it recommends an AI deployment. It checks the business case, operating-model readiness, change enablement and governance exposure, then returns Accelerate, Fix, or Stop with a modelled EUR value.

Assess one AI decisionRead the protocol
DeterministicOpenOn-premMIT
AccelerateFixStop
What it does

Four pillars, one deterministic call

Start with an AI proposal in plain English. The engine resolves the decision inputs, discloses estimated pillars and asks for missing context.

01
Strategic Alignment
How clearly this moves a board-level KPI.
02
Financial Return
Strength of the modelled return.
03
Change Enablement
Sponsor in place, owner named, change budget funded.
04
Governance Risk
Regulatory and reputational exposure. Higher is worse.
GR ≥ 70 or FR ≤ 20 returns Stop. SA, FR and CE ≥ 60 with GR ≤ 40 clear the pillar test for Accelerate. Accelerate also requires evidence that workflows, roles, decision rights and measures are ready. Missing or partial evidence, or a stated gap, holds the verdict at Fix.
The deterministic rule set
MCP Server · 13 tools

Callable from any agent

assess_ai_initiative
Start from a proposal in plain English. Resolve decision inputs, ask for missing context and return a verdict with the work architecture checks.
score_initiative
Four-pillar score → Accelerate / Fix / Stop with EUR value range, decision confidence, applied modules, reasoning.
score_portfolio
Scores every initiative in one call. Returns the board-level shape: verdict counts, aggregate EUR, mean confidence.
assemble_portfolio
Builds the portfolio document from loose inputs: aliases resolved, ids generated, missing pillars estimated with provenance, validated before return.
recommend_improvements
For Stop or Fix, the specific pillar raises that would flip the call toward Accelerate.
calculate_pace_layer_drag
Annual Organisational Drag Cost in EUR from AI-tier vs operating-model misalignment.
validate_portfolio
Validates a portfolio JSON document against the BVF v1.0 schema.
get_benchmark
Disclosed AI BVF planning rates for a business function and industry, with evidence status and use guidance.
list_taxonomy
Valid values for industries, functions, AI tiers, readiness levels.
diagnose_process
Advisor Brain: diagnoses one process from observed signals → verdict + net EUR saving.
sequence_portfolio
Three-wave rollout plan with named gates: Stops first, quick wins second, complex work third, with change capacity enforced per function.
map_to_taxonomy
Everyday business language mapped onto the canonical enums, so callers never guess.
infer_readiness
Measures organisational readiness from process signals (hand-offs, rework, wait time, automation, cycle time) instead of accepting self-report. When the measured answer is lower than the claimed one, the gap is itself a change-readiness finding.
CI/CD pre-flight gate · aibvf-check

SonarQube for AI

Wire the gate into CI. Initiatives that can't survive a board review fail the build, before the budget is committed.

ci · pre-flight gate
$ aibvf-check ./portfolio.bvf.json
Quality gate
min-verdict=Fix · max-governance-risk=60
FAIL
✓CX copilot rolloutAccelerate72
✓HR onboarding assistantAccelerate66
✓Claims triage automationFix54
✕Autonomous procurement agentStop38
✕Contact-centre voice botStop44
What you don't gate, you don't govern.exit 1 · 2 initiatives blocked

Start with one AI decision

Choose a hosted connector or a local MCP client. Both expose the same thirteen tools.

Hosted on claude.ai

In Settings, open Connectors, choose Add custom connector, and paste:

https://mcp.aibvf.com/api/mcp

Local with Claude Desktop, Claude Code or Cursor

Use this configuration in your MCP client, then quit and restart it:

{
  "mcpServers": {
    "aibvf": { "command": "npx", "args": ["-y", "aibvf-mcp"] }
  }
}

Windows needs cmd /c. Open the client setup and troubleshooting guide for platform instructions.

Ask for the first assessment

Assess this AI initiative using AI BVF: we're a EUR 2.4bn manufacturer, planning a GenAI predictive maintenance rollout in our EU plants, we're a traditional hierarchy, strong sponsor, modest change budget.

The agent calls assess_ai_initiative and asks for unresolved decision inputs before returning a verdict. Estimated pillars are disclosed. For a Fix or Stop, ask for recommend_improvements, supply evidence and rerun.