Live tool → https://ogradyliam5.github.io/copilot-studio-credit-estimator/
An interactive, zero-dependency estimator for Microsoft Copilot Studio credit consumption and cost — modeling the Standard harness and the GitHub Copilot harness side by side, because they are not priced the same way. Not even close.
🤖 This tool was developed by Liam's engineering agent, built in the Copilot Studio GitHub Copilot harness running Claude Fable, connected to GitHub through the GitHub MCP server.
Copilot Studio now runs everything on a harness — the runtime between your agent design and the model. The harness you pick changes not just capability, but the entire billing model:
| 🧱 Standard harness | 🐙 GitHub Copilot harness | |
|---|---|---|
| Billing | Fixed per-feature rate card | Usage-based Copilot Credits |
| Authoring | Free until publish | Metered from the first build action (testing & evals billed) |
| M365 Copilot license inclusion | ✅ B2E usage = $0 (fair use) | ❌ Never — all usage billed |
| Typical cost per run | 1–50 credits | 100 to 500+ credits per task |
| Superpowers | Deterministic, auditable, cheap | Agentic loop, file generation, skills, self-recovery |
A light GitHub Copilot harness task (100–300 credits) can cost 5–15× a complete Standard-harness multi-step run — but it can also do things the Standard harness architecturally cannot. This tool lets you put numbers on that trade-off before the invoice does.
- Standard harness rate card — classic answers (1 cr), generative answers (2 cr), agent actions (5 cr), tenant graph grounding (10 cr), agent flow actions (13 cr / 100), AI tools (0.1 / 1.5 / 10 cr per 1K tokens), reasoning-model surcharge (+10 cr / 1K tokens), content processing (8 cr / page)
- Editable rate card — the rate card is split into clearly separated Standard-harness and GitHub-Copilot-harness groups, and every rate can be edited if Microsoft's published rates change. It's locked by default: unlocking requires an explicit confirmation, a warning banner is shown whenever custom rates are in effect, exported reports are flagged as using CUSTOM rates, and a one-click reset restores the list rates
- M365 Copilot license inclusion — enter the number of licensed users (capped at your active users) whose B2E Standard-harness usage bills $0
- GitHub Copilot harness task tiers — light (100–300), medium (300–500), heavy (>500) with a configurable complexity mix and adjustable tier midpoints
- Development burn — GHCP bills authoring, previewing, testing and eval generation; model your makers × test runs × credits per run
- Pack optimizer — cheapest blend of prepaid 25K-credit packs ($200) + PAYG ($0.01/credit) overflow, with a warning when you'd trip the 125% overage enforcement threshold
- Break-even scaling chart, a build-phase vs steady-state chart (what a dev/test month costs before launch vs a typical production month, with the GHCP ongoing dev burn broken out), credit composition breakdown, and a capability-gap matrix for the things money can't buy
- Shareable scenarios — the full input state (including any custom rates) is encoded in the URL; export any estimate as a copy-paste Markdown report
- 📋 Copy Markdown — copies the full report (comparison table, cost multiple, verdict,
assumptions and the scenario link) to the clipboard as Markdown that pastes cleanly
into email, Teams, Word/OneNote, GitHub issues and wikis. If the clipboard isn't available, the
report downloads as a
.mdfile instead.
Every report embeds the scenario link, so anyone reading it can reopen the exact same inputs.
No build step, no dependencies:
git clone https://github.com/ogradyliam5/copilot-studio-credit-estimator.git
cd copilot-studio-credit-estimator
python3 -m http.server 8000 # or just open index.htmlRun the headless smoke tests (validates the calculation engine, the rate-card lock/unlock flow and the report exports against hand-computed expectations):
node smoke-test.js- Billing rates and management (Standard harness) — Microsoft Learn
- Overview of usage-based billing (GitHub Copilot harness) — Microsoft Learn
- Choose a harness — Microsoft Learn
- Quotas and limits — Microsoft Learn
Planning estimate only — not a billing commitment. Rates current as of August 2026 and exclude bring-your-own-model (Azure Foundry) configurations. GHCP tiers are published as ranges; actual consumption depends on agent design, knowledge sources and tool calls. The M365 Copilot inclusion is B2E-only, subject to fair-use limits Microsoft can revise, excludes Computer-Using Agents, and for agent flows applies only to the "When an agent calls the flow" trigger. Always verify against the current Copilot Studio Licensing Guide before procurement decisions.
Developed by Liam's engineering agent — an autonomous software engineering agent built in the Copilot Studio GitHub Copilot harness, running Claude Fable, and connected to GitHub through the GitHub MCP server (which it used to inspect the repository, commit the code and open the pull requests for this project).
MIT