Integration

GenZAgents for GitHub Copilot — signed receipts across every IDE Copilot runs in

GitHub Copilot is the most distributed AI coding tool in the world — VS Code, IntelliJ, PyCharm, GoLand, WebStorm, Rider, Visual Studio 2022, github.com. GenZAgents wires into every surface that supports MCP (Feb 2025+ for VS Code, Apr 2025+ for JetBrains, May 2025+ for Visual Studio) so the audit layer follows your devs regardless of which IDE they prefer.
Install GenZAgents MCPJoin launch waitlist

Why GitHub Copilot is the highest-leverage integration we ship

GitHub Copilot has ~1.8M paid seats — bigger than every other AI coding tool combined. ChatGPT for Work and Claude Pro grow, but Copilot is the one developers use INSIDE the editor, every day, on real code. Capturing the reasoning trail behind Copilot suggestions is therefore the highest-leverage place to apply the GenZAgents audit layer. Every line of code your team merges traces back to a Copilot Chat conversation that drove it. Without GenZAgents, that conversation is gone the moment the dev closes their IDE.

Three IDE surfaces, one MCP config

VS Code Copilot Chat added MCP in Feb 2025 (settings.json `chat.mcp.servers`). JetBrains IDEs followed in Apr 2025 (plugin 1.5.40+, one shared config across IntelliJ / PyCharm / GoLand / WebStorm / CLion / Rider / RubyMine). Visual Studio 2022 added MCP in May 2025 (extension reads %LOCALAPPDATA%\GitHub Copilot Chat\mcp\config.json). @genzagentsio/setup writes all three from one command — no per-IDE setup, no re-config when the dev switches editors mid-project.

.github/copilot-instructions.md is the org-wide hook

GitHub Copilot Chat reads .github/copilot-instructions.md from any repo it operates inside, then prepends the contents to every chat session as a system instruction. This is the canonical place to wire org policy into Copilot — and GenZAgents uses it to tell Copilot Chat to call org_context_lookup BEFORE generating code. Run `npx @genzagentsio/setup --copilot-instructions` in any repo and the file lands with idempotent markers so re-runs update in place rather than appending duplicates. Commit it once, every dev across every IDE inherits the protocol.

The de-dup superpower

Without this protocol: Dev A asks Copilot on Monday to design a Stripe idempotency-key reset; Dev B asks Copilot on Friday for the same thing, gets a similar-but-different design, no one realises. With this protocol: Dev B's prompt triggers org_context_lookup, finds Dev A's receipt from Monday with confidence 0.84, Copilot tells Dev B "@DevA shipped this on 2025-04-15 (rcpt_01H...) — reuse, extend, or supersede?" Compounding effect: every Copilot conversation makes the org slightly smarter than the last one.

What we can capture vs what we can't

CAN capture (via MCP): every tool call the dev makes through GenZAgents tools inside Copilot Chat — draft_receipt, capture_session, org_context_lookup, restore_chat, snapshot_agent_memory. Plus any custom MCP tool you add to your private GenZAgents instance. CANNOT capture: the raw Copilot Chat conversation text without the dev explicitly calling capture_session, Copilot's inline ghost-text suggestions (no public API exists), or Copilot completions in non-chat mode (autocomplete). The .github/copilot-instructions.md protocol mitigates this by routing real work through MCP tools rather than relying on raw chat capture.

Microsoft 365 Copilot (Word / Excel / Outlook / Teams) coming Q3 2026

Microsoft 365 Copilot uses a different extensibility layer (Graph agents + plugins) — not MCP. Capturing those conversations requires tenant-admin consent + Graph audit log access. We're shipping that integration for the Enterprise tier (£499/mo/org) so customers can capture every Copilot Chat conversation across Word, Excel, Outlook, and Teams. Targeting Q3 2026. The receipts pool with the same dashboard so a single org sees BOTH GitHub Copilot and M365 Copilot in one place.

Audit + telemetry parity for Copilot Business / Enterprise admins

If your org runs Copilot Business or Enterprise, GitHub publishes a daily usage report API showing per-developer suggestion counts, acceptance rates, language breakdowns. The GenZAgents admin dashboard pulls this in and overlays it with receipt activity — so org admins see one view of "Copilot suggestions accepted" alongside "receipts issued from Copilot conversations". Useful as a productivity / adoption metric you can show to the CTO who funded Copilot Business but has no way to measure ROI.

Install

$ npx @genzagentsio/setup

Detects every Copilot-enabled IDE on the machine and writes the MCP
config once for all of them. JetBrains IDEs share a single config
at ~/Library/Application Support/github-copilot/intellij/mcp.json
(macOS) — install once, every JetBrains product reads it.

For repo-level priming add the .github/copilot-instructions.md:

$ npx @genzagentsio/setup --copilot-instructions

Drops the file into the repo root. Copilot Chat (every IDE +
github.com) reads it as a system instruction telling Copilot to:
  - Call org_context_lookup BEFORE generating code
  - Call capture_session AFTER substantive answers

Idempotent. Re-run anytime; safe in CI.

What we capture

Every MCP tool call from Copilot Chat in VS Code, JetBrains IDEs, Visual Studio. Every org_context_lookup result (the de-dup signal). Every capture_session (the conversation summary Copilot writes).

Verify it works

In any Copilot Chat: ask "@workspace what receipts exist for this repo" → Copilot calls list_my_receipts via MCP → reply lists your recent receipts. Or: ask Copilot to "use org_context_lookup to check if anyone built X" — the receipt count on https://genzagents.com/dashboard increments by 1.

Common questions

Does this work with the github.com Copilot Chat panel (the one on PR pages)?▾

Yes via the .github/copilot-instructions.md protocol — github.com Copilot Chat reads that file from the repo it's operating on. Direct MCP support for github.com Copilot Chat is on GitHub's roadmap (no public ETA). For now, the instructions file gives you protocol-level coverage even without MCP. We're also building a GitHub Copilot Extension (Marketplace listing) so devs can @-mention `@genzagents` directly in github.com chat — see Copilot-Extension-Plan.md in our roadmap.

I use multiple JetBrains IDEs (IntelliJ + PyCharm + GoLand). Do I install once or per IDE?▾

Once. GitHub Copilot for JetBrains uses a single config at ~/Library/Application Support/github-copilot/intellij/mcp.json (macOS) / %APPDATA%/github-copilot/intellij/mcp.json (Windows). Every JetBrains IDE on the machine reads from the same file. @genzagentsio/setup writes it once and detects all your IDEs.

What about Copilot Workspace (the autonomous coding agent)?▾

Partial coverage. Copilot Workspace exposes a Coding Agent API but it's currently in private preview. When it ships GA we'll register GenZAgents as an agent so every Workspace task fires a receipt. Until then, devs can capture Workspace tasks manually via capture_session in any Copilot Chat surface.

Does the org-wide .github/copilot-instructions.md replace my existing one?▾

No. The installer uses HTML markers (`<!-- GENZAGENTS:BEGIN -->` / `<!-- GENZAGENTS:END -->`) so re-runs update only our block. If your repo already has a copilot-instructions.md, we append our block at the end. If not, we create the file with our block and a `# Copilot instructions` heading you can extend.

My company uses Copilot Business with content exclusions. Do receipts respect those?▾

Yes — receipts only capture what Copilot Chat returns. If your org blocks Copilot from suggesting code based on certain files (Copilot Business content exclusions), those files never appear in Copilot's replies, and therefore never appear in our receipts. Our MCP server operates on the same input Copilot sees. The audit layer doesn't bypass org policy.

How does this compare to the GitHub Copilot Business audit log?▾

Complementary, not competing. GitHub's audit log captures aggregate usage (suggestion counts, acceptance rates per dev). GenZAgents captures the reasoning trail (which architectural decisions Copilot helped with, which design alternatives the dev considered, what got picked and why). One is "how much Copilot did your team use", the other is "what did Copilot help your team decide". Most enterprise customers want both — and the Enterprise tier of GenZAgents pulls in the GitHub audit log so you see both in one dashboard.

Related

Integration: Cursor

Add GenZAgents to Cursor via Settings → MCP. Every Cursor agent action, codebase edit, and Cmd-K session becomes a signed receipt for compliance, cost attribution, and cross-engineer handoff.

Integration: Cline

Wire GenZAgents into Cline's MCP servers list. Every autonomous task run becomes a signed receipt with per-engineer attribution, cost tracking, and full tool-call audit.

Integration: Roo Code

Roo Code (formerly Roo Cline) is the autonomous-agent fork of Cline. GenZAgents captures every Roo agent loop as a signed receipt.

Integration: Continue.dev

Continue.dev is the open-source AI coding assistant. Wire GenZAgents into Continue's MCP config — every assistant call becomes a signed receipt.

Get the trust layer for your AI work

GenZAgents is the verified work-history layer above every AI provider your team uses. Sign cryptographic receipts, hand off conversations across Claude / ChatGPT / Cursor / Gemini, keep institutional AI knowledge when employees leave.

Join the launch waitlistInstall MCP server

Last reviewed · 6 min read· Open spec· Changelog