GenZAgents for Continue.dev — open-source IDE assistant with verified receipts

Continue.dev fills the "open-source assistant" gap

Cursor / Windsurf / Cline are great products, but they're closed-source. Continue.dev is the open-source alternative — VS Code + JetBrains plugin, BYO LLM, fully auditable. Orgs with strict supply-chain compliance (defence, government, finance) frequently mandate open-source AI tools. GenZAgents fits because we're an MCP server: drop in via config, capture starts.

Per-engineer attribution without seat licensing

Continue.dev doesn't require a seat license — engineers install the VS Code extension, configure their own API keys, and go. GenZAgents matches that model: every engineer's laptop has its own human_id in ~/.genzagents-mcp-env, and the receipts are tagged accordingly. No SCIM, no per-seat invoice, just receipts with attribution.

Multi-model awareness

Continue lets engineers pick their model per-conversation (Claude, GPT, local Llama, Gemini). The GenZAgents receipt captures which model was used per call — your receipt feed shows "engineer used claude-3.5-sonnet at 14:32 for the planning, gpt-4o at 14:45 for the implementation". That granularity is exactly what cost-attribution reports need.

JetBrains IDEs (IntelliJ / PyCharm / WebStorm)

Continue ships a JetBrains plugin in addition to VS Code. The MCP server config lives at ~/.continue/config.json regardless of IDE, so the same GenZAgents setup covers both. For a mixed-IDE team (JetBrains for Python, VS Code for TypeScript), this is the only path that captures both consistently.

Self-hosted LLMs (Ollama, vLLM)

Continue.dev supports local model backends like Ollama. The GenZAgents receipt captures the model name + the local runtime ("model=llama-3.1-70b-instruct, runtime=ollama") without us ever seeing the model weights or the inference traffic. This matters for orgs running fully on-prem AI — you get receipts without ever sending data outside your network.

Install for a team via MDM

Same @genzagentsio/setup MDM path as the other IDEs: push the package via JAMF / Intune / Workspace ONE, with the org install token. The setup CLI detects Continue and writes the MCP config. Engineers get the integration on next laptop sync.

Install

Edit ~/.continue/config.json:

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

What we capture

Every Continue assistant call that touches an MCP tool.

Verify it works

Open Continue → ask the assistant to call the genzagents draft tool → confirm receipt on /dashboard.

Common questions

Does GenZAgents capture Continue's autocomplete?▾

No — autocomplete in Continue doesn't invoke MCP tools, so we don't see it. The chat + agent modes do invoke MCP, so those are captured. If you need autocomplete coverage, talk to us about a Continue-specific completion-event adapter (v0.8).

What if my engineers self-host Continue Hub?▾

No difference. GenZAgents sits in the engineer's laptop config, not in Continue Hub. The capture works regardless of whether the chat backend is hosted or self-hosted.

Can I use Continue's context providers + GenZAgents together?▾

Yes. Continue's context providers (@codebase, @docs, etc.) feed input to the model; GenZAgents sits at the MCP tool boundary and captures the output side. The two are orthogonal.

Do receipts work when Continue is in "ask" mode vs "agent" mode?▾

Both. Ask mode produces single-turn receipts; agent mode produces multi-step receipts with parent_run_id linking the sub-steps. The receipt schema handles both.