# Augustus

> Augustus is the AI operating layer for a business. It audits a repetitive process, builds an AI system for it, tests it on the business's own real data, and runs it with a human approving what matters. Augustus is MCP-native: you drive it from your coding agent, with no dashboard.

Augustus takes the repetitive, rule-governed work a business runs on (invoice follow-up, intake and onboarding, document chasing, reconciliation, approvals, renewals, triage, and more) off the team's plate, and comes to the owner only for the decisions that matter. Building and testing are free; it is $199/month per seat once a system goes live. No sales calls, no usage fees, no per-task pricing.

## Connect (MCP)

- MCP endpoint: https://app.augustus.so/api/mcp (HTTP transport)
- Add to a coding agent: `claude mcp add --transport http augustus https://app.augustus.so/api/mcp`
- No coding tool needed: in Claude.ai, Settings, Connectors, Add custom connector, paste https://app.augustus.so/api/mcp, then sign in (OAuth).
- Auth: OAuth in the browser, or an instant key from https://app.augustus.so/api/instant-key
- Server card: https://augustus.so/.well-known/mcp.json
- On connect, follow the Augustus protocol in order. Each brief tool returns its own methodology.

## How it works

- Audit: Augustus learns the business and quantifies what a process costs today, grounded only in real facts (it refuses to build until the critical facts are known).
- Build: it designs the system as a human-in-the-loop workflow graph, using only the tools that are actually connected.
- Test: it runs on the business's own real data in safe preview and must be graded correct by a human before it can go live.
- Run: live systems run on their triggers; irreversible actions wait for one-tap human approval by email. No dashboard to operate.

## The protocol (in order)

1. Foundation: set the business profile from the company website and any uploaded documents.
2. Audit: write the audit on your own model from audit_brief, submit_audit; Augustus validates and grounds it.
3. Ground: answer the blocking questions until the audit is grounded.
4. Build: design the workflow from build_brief, submit_build; it enters safe testing.
5. Test and verify: test_system on real sample data, grade_run, until it is ready.
6. Go live: go_live (passes its acceptance test, then a card), and wire the trigger.
7. Operate: approvals, escalations, and revisions, all human-in-the-loop.

## Links

- Skills (the tools an agent can drive): https://augustus.so/skills
- Docs and the open protocol: https://github.com/augustus-so/protocol
- Knowledge (AI in plain words): https://augustus.so/knowledge
- Run Augustus for your clients: https://augustus.so/partners
- Pricing: $199/month per seat, flat. Free to sign up, audit, build, and test.