Integration

GenZAgents for Cline — receipts for autonomous coding agents

Cline runs autonomous agent loops that touch your filesystem. GenZAgents turns each loop into a signed receipt so you can audit, attribute cost, and replay any session a regulator asks about.

The autonomous-agent audit gap

Cline is one of the most aggressive autonomous-loop products: it will read, write, and execute on your filesystem with minimal hand-holding. The audit gap is correspondingly larger — "what did the agent do for the last 45 minutes" is a serious compliance question. GenZAgents fills it with per-loop receipts that capture tool calls, approval decisions (yes/no/abort), and the runtime cost. The receipt is the contract: it's what your security team reviews after the fact.

Approve-step granularity

Cline's default policy asks for human approval before every potentially-destructive action. GenZAgents captures each approve/deny as a sub-receipt field — so the audit log says "engineer approved write to src/auth.ts at 14:32" not just "engineer ran a session at 14:30". This is the granularity a SOC 2 control review actually wants.

Receipt + run resumption

Cline supports task resumption — pause a long agent loop, come back tomorrow. GenZAgents tracks resumption: the new receipts include a parent_run_id linking back to the original run. The dashboard shows the full timeline including the pause + resume gap, which matters when a regulator asks "what was the timeline of the AI-assisted incident response?"

Cost attribution per autonomous run

Autonomous loops eat tokens. Without per-run cost attribution, your AI spend report is a single mysterious line item. With GenZAgents: each receipt carries the runtime cost; the dashboard rolls them up by engineer + project; the monthly report breaks down "$X on Cline this month, of which $Y was the auth-service migration project, of which $Z was Bob's runs".

Cross-IDE handoff

A senior engineer might prototype with Cline (autonomous loops) and finalise with Cursor (fast inline edits). GenZAgents unifies the receipts: same agent DID, same project tag, both IDEs feed into the same receipt timeline. The cross-IDE handoff is invisible to the dashboard — the only thing that matters is the agent + the project, not which editor was open.

Install via MDM for a team

Cline is increasingly common in security-conscious orgs because it's open-source. The flip side: there's no app-store-style mass install. Use @genzagentsio/setup via your MDM with the org install token. The setup CLI detects Cline's settings.json and writes the GenZAgents MCP block. Engineers wake up with the integration already wired in.

Install

Cline → Settings (gear icon) → MCP Servers → Edit Settings → add:

{
  "mcpServers": {
    "genzagents": {
      "command": "npx",
      "args": ["-y", "@genzagentsio/mcp-server"]
    }
  }
}

What we capture

Every Cline autonomous task, every MCP tool invocation, every approve-and-continue step.

Verify it works

Start a Cline task → wait for the first MCP tool call → see the receipt land on /dashboard within 3 seconds.

Common questions

Does GenZAgents capture Cline's "execute command" tool calls?

Yes — execute_command is just another MCP tool from the receipt's perspective. We capture the command string + the exit code + the timestamp. The captured stdout is digest-only by default; flip GENZAGENTS_CAPTURE_STDOUT=true to include the full output.

What about Cline's built-in browser-use mode?

Captured. Every browser action (click / type / navigate) becomes a sub-event on the parent receipt. The full DOM snapshot at each step is digest-only by default; turn it on with GENZAGENTS_CAPTURE_DOM=true (storage cost is non-trivial — pick which agents need it).

Can I run Cline + GenZAgents without internet?

Not today. The MCP server requires a network round-trip to api.genzagents.com to issue receipts. v0.8 ships an offline-queue mode where receipts buffer locally and flush on next online check; in the meantime, the receipts fail closed (Cline still works, just no receipt).

Does Cline's plan mode produce a receipt?

Yes — plan mode produces a "plan_drafted" receipt with the proposed steps. When the user approves and execution begins, a "plan_executed" receipt is emitted with parent_run_id pointing at the plan_drafted one.

Related

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.

Last reviewed · 3 min read· Open spec· Changelog