Aibvf MCP

What aibvf-mcp does

The aibvf-mcp MCP server applies the AI BVF planning model to individual AI initiatives and portfolios. An initiative assessment produces a Stop, Fix, or Accelerate verdict, identifies missing evidence, shows estimated or supplied inputs, and recommends the next review action. The model evaluates four pillars: Strategic Alignment, Financial Return, Change Enablement, and Governance Risk.

Portfolio workflows support assembly, schema validation, scoring, and sequencing. The portfolio tools can organize loose inputs, expose unresolved terms, calculate an aggregate shape, and produce rollout waves constrained by change capacity. Any aggregate benefit range still requires a separate review for overlapping work and shared benefits.

How it works

The aibvf-mcp MCP server accepts either a plain-language proposal or explicit structured inputs. assess_ai_initiative resolves the proposal, requests missing decision inputs, and returns an assessment. score_initiative evaluates explicit values, applies the work-architecture gate, and returns reasoning, audit information, and sensitivity details. Identical inputs produce identical outputs for a given engine version.

The decision rules include Stop conditions at Governance Risk of 70 or higher, or Financial Return of 20 or lower. Accelerate requires Strategic Alignment, Financial Return, and Change Enablement of at least 60, Governance Risk of no more than 40, and all four work-architecture checks. Missing evidence, partial assessment, or incomplete architecture evidence can keep an otherwise positive initiative at Fix.

The decision score and confidence-related fields are rule-based summaries. They do not represent a measured probability of success or prediction accuracy. Planning benefit values use a readiness-adjusted revenue scenario and do not include project costs, margins, timing, or overlap.

Setup and configuration

For local use, add the package to an MCP client with npx:

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

Restart the client after adding the configuration. Windows requires cmd /c around the command. A hosted Streamable HTTP endpoint is available at https://mcp.aibvf.com/api/mcp for a custom connector on claude.ai.

Tools and capabilities

The package exposes 13 tools:

• Assess and score initiatives with assess_ai_initiative and score_initiative.
• Suggest pillar actions and work redesign through recommend_improvements.
• Assemble, validate, score, and sequence portfolios with the corresponding portfolio tools.
• Diagnose process signals and infer readiness with diagnose_process and infer_readiness.
• Model pace-layer drag and retrieve planning rates with calculate_pace_layer_drag and get_benchmark.
• Inspect accepted categories and resolve business terms using list_taxonomy and map_to_taxonomy.

Limitations and notes

The aibvf-mcp MCP server records supplied values and architecture checks, but teams remain responsible for reviewing the evidence. Research informs the questions; it does not validate the model's rates, multipliers, readiness percentages, or drag rates. Module labels do not perform clinical validation or certify regulatory compliance.

Optional anonymous telemetry can report tool calls and connection events. It excludes portfolio content, revenue figures, numeric pillar scores, and personal identifiers. Set AIBVF_TELEMETRY_DISABLE=1 to disable events and local install-ID creation. AIBVF_TELEMETRY_URL and AIBVF_TELEMETRY_KEY can direct telemetry to another backend. Review the formulas and worked example before using results in funding decisions.