Power your agent
The publisher's own URL. Not a PlugMCP endpoint.
Host a copy with our keys and logs.
https://mcp.aibvf.com/api/mcpThe Claude button above opens Claude with the name and URL filled in. You still confirm it there.
claude mcp add --transport http io.github.craig-horton-aibvf-mcp https://mcp.aibvf.com/api/mcpRun in a terminal.
{ "mcpServers": { "io.github.craig-horton-aibvf-mcp": { "command": "npx", "args": [ "-y", "mcp-remote", "https://mcp.aibvf.com/api/mcp" ] } }}claude_desktop_config.json
Remote URL or a command on your machine? How the two installs differ
Run on PlugMCP
A new server you create. It is not this listing's official endpoint. Use the publisher install when you want their URL and their credentials.
LaunchFrom the publisher
Turn an AI proposal into a decision brief: the verdict, the evidence gaps and the next action for the person who owns the work.
Open aibvf.com/start. Your first assessment needs no account or installation.
Review an AI proposal and return Stop, Fix or Accelerate, the decision gaps and the next action.
Paste this synthetic example, or describe a proposal you are reviewing:
We are a EUR 2.4bn manufacturer considering GenAI for predictive maintenance in our supply chain. Our operating model is traditional. We have a sponsor, but the affected roles, human override rights and performance measures still need review.
Run the assessment, check how the proposal was interpreted, and correct the assumptions. The assessment asks for unresolved inputs and marks estimated pillar scores. Keep the action list, name an owner and return with evidence when the work changes.
The result brings the following questions together. This is an illustrative reading guide for the example, with the engine's field names documented below.
For the example above, the first useful action is to evidence the proposed workflow and its owners. An Accelerate verdict requires the pillar thresholds to clear and all four work architecture checks to be met.
See the reproducible worked example for exact inputs, calculations and a re-score that stays at Fix until the work architecture is evidenced.
The browser is the first-use route. The hosted connector and local MCP package let an assistant repeat the assessment while you work on a proposal.
Open Settings, then Connectors, then Add custom connector, and paste:
https://mcp.aibvf.com/api/mcp
Then ask:
Assess this AI initiative using AI BVF: we are a EUR 2.4bn manufacturer considering GenAI predictive maintenance, with a traditional operating model. Resolve the inputs, show the assumptions and identify the evidence needed for the next decision.
Add this configuration to your MCP client:
{
"mcpServers": {
"aibvf": { "command": "npx", "args": ["-y", "aibvf-mcp"] }
}
}
Quit and restart the client. Windows uses cmd /c; the client setup and troubleshooting guide contains the complete configuration.
Start with assess_ai_initiative. Use recommend_improvements for a Fix or Stop, review the proposed actions with the people who own the work, and re-score after the evidence changes.
AI BVF is a deterministic planning model with disclosed assumptions. The source and formulas can be inspected, and identical inputs produce identical outputs for a given engine version.
The four pillars are Strategic Alignment, Financial Return, Change Enablement and Governance Risk. GR >= 70 or FR <= 20 returns Stop; SA >= 60, FR >= 60, CE >= 60 and GR <= 40 clear the pillar test for Accelerate. A gap, partial assessment or missing work architecture evidence holds an otherwise green initiative at Fix.
Keep these boundaries with the result:
confidence, decision_confidence and projected_confidence retain their names for compatibility. These are rule-based scores, with no measured probability of project success or prediction accuracy.net_low_eur, net_high_eur and MCP net_value_eur apply a readiness capture assumption to a revenue-based benefit scenario. Project costs, margins, timing and overlap are outside that calculation.applied_modules records implementation context. Labels such as healthcare_clinical_validation and financial_dora_module do not perform clinical validation or certify regulatory compliance.Read the scoring formulas and worked example before using the outputs in a funding decision.
Thirteen tools are exposed through local stdio and the hosted Streamable HTTP connector.
For portfolios, use assemble_portfolio, validate_portfolio, score_portfolio, then sequence_portfolio. An aggregate modelled range needs a separate review of overlapping work and shared benefits.
The public portfolio specification is version 1.0. That document format has a separate version from the packages implementing it; the package version identifies the code and behaviour used for a particular assessment.
The MCP server can report tool calls and a server_connect event. Events include protocol and package versions, entry route, assessment stage, work architecture status, taxonomy fields, a daily-rotated caller hash, and classification plus confidence where supplied.
Local stdio calls also include a stable one-way install_hash for repeat-use measurement, derived from a random local seed. Hosted calls send no stable install hash, and a broad user_role is sent only when a local user explicitly sets AIBVF_USAGE_ROLE.
No portfolio content, revenue figures, numeric pillar scores or personal identifiers are included. Set AIBVF_TELEMETRY_DISABLE=1 to prevent events and creation of the local install-id file. Point at your own backend with AIBVF_TELEMETRY_URL and AIBVF_TELEMETRY_KEY.
Package downloads include repeat installs, dependencies and automation. Use completed assessments and subsequent decision reviews to evaluate adoption.
Bring a reproducible counterexample: the inputs, actual output, expected decision, supporting evidence and engine version. The contribution guide includes a template and explains the review and licensing boundaries.
The ten-team pilot pack defines the first-use and return-use checks, interview prompts and tracker. Examples in this repository are synthetic unless a case explicitly records consent and its evidence.
If AI BVF helped you review a decision, star the repository or share a counterexample. Both give the project useful feedback.
Repository source code is MIT licensed under LICENSE. The specification and JSON Schema under spec/ are CC-BY-4.0, and the AI BVF names and logo are trademarks, as set out in NOTICE.
Private benchmark material and certification marks are outside the source-code contribution route. The contribution guide explains how to discuss those materials without changing the rights granted by the repository licenses.
Craig Horton is an independent transformation lead based in Amsterdam and the author of the AI Business Value Framework. His work connects AI investment decisions with organisational readiness and the redesign of work.
© 2026 MCP Directory.
Review an AI proposal and return Stop, Fix or Accelerate, the decision gaps and the next action.
{ "mcpServers": { "io.github.craig-horton-aibvf-mcp": { "url": "https://mcp.aibvf.com/api/mcp", "headers": {} } }}Tools
No tool list yet. Remote servers get one after the first health check.
FAQ
Copy the http config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.
No required environment variable is listed on this record.
AI Business Value Framework is reached over HTTP.
Last checked 2026-11-15. Review an AI proposal and return Stop, Fix or Accelerate, the decision gaps and the next action.