Local AI development

Connect a local AI coding agent to a running d6e instance and develop/test Plugins and workspace assets — no special deploy step until you are ready to ship.

The core idea: everything the hosted AI agent can do is available as a public HTTP API / MCP. Your laptop gets the same ~96 d6e_* tools the built-in chat uses.

flowchart LR
  agent["Local AI agent<br/>Cursor / Claude Code / Codex"]
  inst["d6e instance<br/>MCP :8081/mcp → Rust /api/v1/* → PostgreSQL"]

  agent -->|"MCP"| inst
  agent -->|"REST"| inst

Text version (ASCII)

┌────────────────────┐        ┌────────────────────────────────────┐
│ Local AI agent     │  MCP   │ d6e instance                       │
│ (Cursor / Claude   │───────▶│  MCP :8081/mcp  →  Rust /api/v1/*  │
│  Code / Codex)     │  REST  │                 →  PostgreSQL      │
└────────────────────┘───────▶└────────────────────────────────────┘

This page is a summary. Full steps, differences, and troubleshooting live in the source guide:

Canonical: local-ai-development.md

0. Install Agent Skills (recommended)

Before asking an agent to design, answer questions, or implement Plugins / Docker STFs / custom frontends, install the skills. When unsure, install everything.

npx skills add d6e-ai/d6e-plugin-skills --skill '*' -y
npx skills add d6e-ai/d6e-docker-stf-skills --skill '*' -y
npx skills add d6e-ai/d6e-custom-frontend-skills --skill '*' -y

See Installing Agent Skills for details.

1. Issue an API key (~1 minute)

Log into the console (${D6E_BASE_URL})
Avatar → API keys (/{locale}/user/api-keys)
Create a key and copy the d6e_… value (shown once)

Smoke test:

curl -s ${D6E_BASE_URL}/api/v1/workspaces \
  -H "Authorization: Bearer ${D6E_API_KEY}"

API keys are tied to your user account and inherit workspace membership. Add X-Workspace-ID for workspace-scoped endpoints.

2. Connect MCP (recommended)

The instance exposes MCP over HTTP (default port 8081, path /mcp).

Cursor — .cursor/mcp.json:

{
  "mcpServers": {
    "d6e": {
      "url": "http://<instance-host>:8081/mcp",
      "headers": { "Authorization": "Bearer d6e_YOUR_API_KEY" }
    }
  }
}

After connecting, you can ask the agent to:

“List tables in the current workspace”
“Run this SQL and summarize the result”
“Instant-run this STF”

See MCP tools for the full catalog.

3. What to test where

What you want to test	Where it runs	How
Workspace SQL	Remote (real DB)	`POST .../sql` or `d6e_sql`
JS STF	Remote (QuickJS)	`POST .../stfs/instant-run` or `d6e_instant_run_stf`
Docker STF logic	Local `docker run`	stdin/stdout JSON; point `api_url` at the live instance for integration
SaaS calls	Remote	`POST .../saas-proxy` or `d6e_call_external_api`
Workflows	Remote	`POST .../workflows/{id}/execute` or `d6e_execute_workflow`

4. Plugin vs frontend scope

This guide: developing and testing workspace contents (Plugins)
Custom frontend: an independent web app — see Custom frontend and frontend-and-instance.md