mova-flat-runner

Public test release — lightweight contract execution for Claude and MCP-compatible AI clients.

MOVA Flat Runner is a small, practical facade for running AI-assisted business workflows with human approval and an audit trail.

It gives Claude a contract-shaped way to handle bounded workflows such as invoice review, supplier screening, complaint triage, refund approval, purchase order review, or internal approval tasks.

This is intentionally not the full MOVA State engine.

You do not need the broader MOVA ecosystem, a state machine setup, or a custom backend to try it. The goal of this repository is simple:

connect MCP client
→ run a contract-shaped workflow
→ keep a human decision gate
→ receive an audit trail

This repository is currently in a public feedback phase.

A temporary shared API key is provided so the Anthropic / MCP community can try the runner with minimal setup. After feedback is collected, the test API will be replaced by a production implementation with proper accounts, API keys, quotas, private connector configuration, and clearer deployment boundaries.

Use this release for testing, prototyping, and feedback.

Do not send sensitive production data, regulated customer data, secrets, credentials, private keys, or confidential business documents through the shared test API.

What it does

MOVA Flat Runner lets an AI assistant run a bounded workflow instead of improvising.

It can:

• run built-in demo workflows
• run registered custom contracts
• validate inputs before execution
• call a configured LLM provider for analysis steps
• keep a human decision gate before final outcome
• record an audit trail for each run
• work from Claude Desktop, Claude Code, Cursor, or other MCP-compatible clients

The core idea:

AI proposes analysis
contract constrains the workflow
human approves the final decision
audit records what happened

What it is not

This public test release is:

• not a general autonomous agent
• not a replacement for Claude Skills
• not the full MOVA State 1.5 engine
• not a certified compliance, credit, legal, medical, or financial decision system
• not production infrastructure
• not a place to send sensitive data through the shared API key
• not a system that makes final business decisions without a human gate

The runner is intentionally flat and practical. For high-risk, multi-step, policy-heavy execution with formal transition authority, the broader MOVA State engine is a separate layer.

Current release status

The shared API key may be rate-limited, rotated, or disabled after the feedback phase.

Production plans include individual API keys, account-scoped contract storage, quotas, private connector configuration, stronger audit export, clearer retention policy, and an optional private execution path for sensitive workflows.

Temporary test API access

For this public test release, use:

MOVA_API_KEY=test-key-001

This key is shared and temporary. Use it only for testing, prototyping, and feedback — not for real customer data, regulated decisions, confidential documents, credentials, or production workflows.

Use synthetic examples, public test files, or non-sensitive documents.

60-second test

After setup, ask Claude:

Use MOVA health check.

Then try the invoice example:

Process this invoice: https://raw.githubusercontent.com/mova-compact/mova-flat-runner/main/test_invoice_INV-2026-0441.png

Expected flow:

1. MOVA starts an invoice contract
2. The invoice is analysed
3. Verification findings are returned
4. Claude asks you for a human decision
5. You choose: approve / reject / escalate
6. MOVA returns an audit receipt

Built-in d

... View full README on GitHub