Scope and cost-estimate any AI agent build in seconds, right inside Cursor.
A Cursor marketplace plugin by BMD PAT LLC.
Describe the agent you want to build in plain English. Agent Architect returns a formatted scope report with:
- Complexity tier — DIY / Startup / Growth / Enterprise
- Cost breakdown — monthly API costs, infra costs, dev hours, total build cost
- Architecture — orchestration pattern, tools, memory approach, key decisions
- Timeline — weeks-to-launch range
- Top 3 risks for this specific agent
The Cursor agent does the work — no extra API keys, no separate process.
Search for Agent Architect in the Cursor Marketplace and click Install.
Clone this repo, then symlink or copy into Cursor's plugins directory:
# macOS / Linux
ln -s "$(pwd)/agent-architect" ~/.cursor/plugins/agent-architect
# Windows (PowerShell, run as admin)
New-Item -ItemType SymbolicLink -Path "$env:USERPROFILE\.cursor\plugins\agent-architect" -Target "$(Get-Location)\agent-architect"Restart Cursor. The skill will be auto-discovered.
In Cursor's Agent chat:
-
Auto-trigger — describe what you want and ask for a scope:
Scope an agent that reads my unread Gmail every morning, drafts replies for the routine ones, and posts a summary to Slack.
-
Manual — type
/and search forscope-agent, then paste your description.
You'll get back a Markdown scope report.
5 scopes per calendar month, tracked locally at ~/.agent-architect/usage.json. The counter resets on the 1st of each month.
Heavy user? Add runtime rails on the agents you ship: AgentGuard. pip install agentguard47
agent-architect/
├── .cursor-plugin/
│ └── plugin.json # Plugin manifest
├── skills/
│ └── scope-agent/
│ └── SKILL.md # The skill (frontmatter + instructions)
└── README.md
Two files do all the work: .cursor-plugin/plugin.json (manifest) and skills/scope-agent/SKILL.md (the skill itself).
That's the entire plugin. No build step, no Node dependencies, no compiled code.
| Tier | Build cost | Timeline | Profile |
|---|---|---|---|
| DIY | $500–$2K | 1–2 weeks | Single-purpose, minimal state, weekend build |
| Startup | $2K–$15K | 2–6 weeks | Multi-step, tool use, basic memory |
| Growth | $15K–$50K | 6–16 weeks | Multi-agent, complex state, production reliability |
| Enterprise | $50K–$150K | 16+ weeks | Mission-critical, multi-system, compliance |
Almost every scope comes back with cost runaway or tool-call loops in the top 3 risks. That is a runtime problem, not a scoping problem.
AgentGuard is a Python SDK that enforces budget, loop, timeout, and rate limits at runtime. Drop it in, set the rails, ship to production.
pip install agentguard47MIT