The agent MCP

Your agent, wired
into your build

One paste and your coding agent stops guessing. It pulls the phase you are on, ships it, and marks it done, so your dashboard is always the truth and you never re-explain the project again.

One paste, then it just knows

Drop the config into your agent's MCP settings and restart it. From then on, "continue my build" is a complete instruction: it reads the current phase, does the work, and marks that phase shipped before it moves on.

.mcp.json
{
  "mcpServers": {
    "upgradiq": {
      "url": "https://mcp.upgradiq.com",
      "headers": {
        "Authorization": "Bearer YOUR_TOKEN"
      }
    }
  }
}
Your token is watermarked to your seat

Three tools, deliberately narrow

list_active_builds

Lists the builds you have started. It cannot see, search, or enumerate the rest of the catalog.

get_current_phase

Returns the one phase you are on, plus the ones you finished. Never the ones that come next.

get_idea_context

Serves the validated research behind your build, so the code comes out market-correct first time.

mark_phase_complete

Your agent marks the phase shipped, and your dashboard updates the moment it does.

report_metrics

Takes the numbers your agent read from your project locally. Numbers arrive; keys never move.

report_blocker

Files what the build is stuck on, so your dashboard can surface the right next action for you.

No list_all, no export, no enumerate. Narrow is the security model.

What we never touch

Your keys stay in your env

We never ask for them, never receive them, and never store them. Nothing to leak from our side.

Only numbers travel

Your agent can report progress and metrics. It cannot send us your code, and we do not want it.

Scoped by design

The tools return your active build and current phase. There is no fetch-all, so a leaked token can never be used to drain the catalog of ideas.

Revocable and watermarked

Every token is tied to your seat and watermarked, so a leak is traceable and killable in one click.

Not every number is equal

An agent can report anything. So reports are tiered by how they were obtained, and the weak tier never gets to matter.

Verified

Read from your real project by your agent, or checked by us on your live URL. This is the evidence that our whole ship guarantee actually runs on.

Observed

What you actually did inside UpgradIQ: phases checked, sessions run. We record it ourselves.

Self-reported

A number you typed. Shown to you only, never compared to anyone, never feeding a benchmark.

If your plan lapses, the token locks

No grace hacks, no quiet draining. The connection stops the moment the plan does, and everything you already built stays exactly where it is on your dashboard, still yours to come back to whenever you want.

See pricing

Frequently asked

Which agents does it work with?

Anything that speaks MCP. Claude Code and Cursor are the primary targets and take a single config block.

Do you see my code?

No. The MCP hands your agent context and takes back progress. Your code never leaves your machine.

Could a leaked token drain the catalog?

No. The tools are scoped to your active build and current phase, with per-token rate limits and no enumerate or fetch-all.

What happens if my plan lapses?

The token locks automatically and your agent disconnects. Your saved build progress stays untouched on your dashboard.

Can my agent lie about progress?

It can report, but reports are tiered. Self-reported numbers never feed a benchmark or the guarantee, which we check ourselves.

Is the MCP included?

It is part of Pro and Max. The free tier keeps the dashboard, without the agent connection.

Pro and Max

Stop re-explaining your project

Wire the agent once, and every session starts where the last one ended.

Your keys never leave your machine · token revocable anytime