upload-artifact

Upload Artifact

Upload file content as a ClosedLoop MCP artifact. Two modes:

1. Script mode (preferred) — uses a standalone Python script that reads the file and calls MCP directly over Streamable HTTP. No conversation context consumed for file content. Requires CLOSEDLOOP_API_KEY and NEXT_PUBLIC_MCP_SERVER_URL to already be present in the current shell environment.

2. MCP fallback — reads the file into context and calls mcp__closedloop__create-document directly. Uses Claude Code's existing MCP auth. Used when the required script-mode environment variables are not available.

Workflow

Follow these steps in order:

Step 1: Resolve Credentials and Choose Mode

Read these values from the current shell environment. Do not read, source, or otherwise rely on a .env.local file in the current working directory:

• CLOSEDLOOP_API_KEY — the API key (starts with sk_live_)
• NEXT_PUBLIC_MCP_SERVER_URL — the MCP server URL

If both exist → use script mode (Steps 2a–5a). If either variable is missing → use MCP fallback (Steps 2b–5b).

---

Script Mode (required env vars available)

Step 2a: List Projects

Run the script with --list-projects:

uv run --with 'mcp[cli]>=2,<3' ${CLAUDE_SKILL_DIR}/scripts/upload_artifact.py \
  --url "$NEXT_PUBLIC_MCP_SERVER_URL" \
  --api-key "$CLOSEDLOOP_API_KEY" \
  --list-projects

Parse the JSON output. Extract the items array. Each item has id and name.

Use AskUserQuestion to present the projects to the user:

• Question: "Which project should this artifact be uploaded to?"
• Options: one per project, label = project name, description = project ID

If only one project exists, skip the question and use it automatically.

Step 3a: Collect Remaining Parameters

Use AskUserQuestion to collect any parameters the user hasn't already specified:

• file_path: "Which file should be uploaded?"
• title: "What title should this document have?"
• type: "What type of document?" (options: PRD, IMPLEMENTATION_PLAN, FEATURE, TEMPLATE)