Turn VS Code agent output into a link

Copilot's agent mode in VS Code produces the same things every coding agent does: implementation plans, change summaries, migration checklists, an explanation of a module for whoever joins next week. All of it lands in the Chat view, where it scrolls away. MintPage gives the agent a publish_page tool, so the same output becomes a URL you can drop into a pull request, a Slack thread or a ticket.

Connect VS Code

VS Code reads MCP servers from .vscode/mcp.json in your workspace (shared with the repo) or from your user profile (every workspace). The VS Code format uses a top-level servers object, not mcpServers:

{
  "inputs": [
    {
      "type": "promptString",
      "id": "mintpage-key",
      "description": "MintPage API key (mp_…)",
      "password": true
    }
  ],
  "servers": {
    "mintpage": {
      "type": "http",
      "url": "https://mintpage.ai/mcp",
      "headers": { "Authorization": "Bearer ${input:mintpage-key}" }
    }
  }
}

The inputs block means VS Code asks for the key once, stores it, and the file you commit never contains it. The same server from the command line:

code --add-mcp '{"name":"mintpage","type":"http","url":"https://mintpage.ai/mcp"}'

Or run MCP: Add Server from the Command Palette and pick HTTP. Start it with MCP: List Servers, open Copilot Chat in agent mode, check that publish_page appears under the tools icon, then ask: "publish this plan as a page and give me the link."

Leave out inputs and headers to use guest mode: no key, no account, pages last 24 hours. A free key from /start makes them last 7 days and unlocks stats; Pro ($5/month or $39/year) makes them permanent with custom slugs.

Good uses

• Plan before edit. Ask the agent for a plan, publish it, and get a reviewer's sign-off before anything touches the repo. The agent can update_page at the same URL as the plan evolves.
• Pull request companions. A long "what changed and why" is better as a link in the PR description than as a 400-line comment.
• Module explainers. "Explain how authentication works in this codebase, then publish it" gives a new teammate a page instead of a chat transcript they will never find again.
• Status pages that stay current. Publish once, then let a task or an agent call update_page after every run.

Example call

{"name":"publish_page","arguments":{"title":"Auth refactor — plan v2","content":"# Plan\n\n1. Extract token validation into `auth/verify.ts`\n2. Add refresh-token rotation\n3. Remove the legacy session middleware\n\n**Open question:** keep the 30-day session cap?","theme":"ink","byline":"Drafted by Copilot in VS Code"}}

What stays safe

Pages are public to anyone with the link, and the tool description tells the agent never to publish secrets. HTML is sanitized (no scripts, forms or iframes) and served with a strict Content-Security-Policy. Delete a page any time with delete_page.

FAQ

Workspace or user profile? Put it in .vscode/mcp.json when the whole team should get the tool with the repo; put it in your user profile (MCP: Open User Configuration) when you want it in every project. Keys belong in inputs either way, never in the committed file.

Does it need OAuth? No. MintPage is a plain streamable-HTTP MCP server: type: "http", a URL and an optional bearer header is the whole configuration.

Can the agent update a page instead of making a new one? Yes: update_page keeps the URL and replaces the content.

Is there a size limit? Guest pages up to 250 KB, free keys 1 MB, Pro 2 MB. Markdown and sanitized HTML are both accepted.

Which tools does the server expose? publish_page, update_page, list_pages, delete_page, get_page_stats, get_account and rotate_api_key. MCP: List Servers → Show Output lists them after the server starts.

Other ways in

The same server works from agent reports, ChatGPT, Claude.ai & Desktop, Claude Code, Codex CLI, Cursor, Gemini CLI, the Markdown API, Windsurf. Full tool reference in the docs; live examples on the showcase.