---
title: "MCP Questions and Recipes"
canonical_url: "https://usepowerplant.com/docs/mcp-recipes/"
md_url: "https://usepowerplant.com/docs/mcp-recipes.md"
last_updated: "2026-10-27"
---

# MCP Questions and Recipes

While the [dashboard](https://app.usepowerplant.com/) gives you an overview of some of Powerplant's capabilities, the MCP is where you put it to work.

## The MCP

The Powerplant MCP is by far the most powerful part of Powerplant. From your own agent, you can ask questions and start work informed by how the whole team is working. Each prompt below can be pasted into Claude Code, Codex, or Cursor as is.

```demo
I'd like to test what the Powerplant MCP can do for our team. First, check that we can query the MCP from here, and that multiple team members have synced more than 5 sessions that we can see.

If so, start each of these prompts in a sub-session, and use the Powerplant MCP to help answer the questions.

1. What were our most expensive PRs in the last month? Are there any patterns in terms of developer and agent friction that contributed to these being more expensive than they needed to be? Any loops where an agent was wasting tokens because it wasn't configured to be able to verify its own work?
2. Prep a "My work this week" recap from my last 7 days' sessions, focusing on my highest impact work.
3. Ask me about some recent bug that was probably caused by a coding agent. Once I reply, use Powerplant MCP and git blame to find the source session that contributed the bug. Was the risk flagged by the agent or developer in that session? Did the presence or absence of certain guidance in our docs contribute to it?
4. What are the most-used skills on our team? How many people use each, and how regularly are they being improved?
```

## Improve PR velocity

```text
What are some of the most common reasons a reviewer has rejected or pushed back on larger, agent-generated PRs in the last 4 weeks? Are there "low hanging fruit" guidelines, lints, deterministic checks, or PR formatting skill instructions that could help here? Rank from likely highest ROI on down.
```

```text
What are the most common corrections or pushback that the team has given agents in the last 4 weeks? Which of these, if any, have guidelines in AGENTS.md that should or could have helped agents not make these mistakes? Are there short, simple principles or deterministic checks we could add that would help the entire team in this regard?
```

```text
To prep me for reviewing PR 9938, take a look in Powerplant for the prompts that Steve used to build this. What testing has been done so far, and does it indicate he weighed the tradeoffs of adding this dependency vs. rolling our own?
```

## Improve agent skills

```text
What skills are the team using most? Which of those skills are candidates for helping improve PR velocity, by reducing the risk of merging or making review easier?
```

```text
Are there any skills people are using that might be better shared in-repo, so they can be iterated on and improved as a team?
```

## Reduce token cost

```text
Of the most expensive PRs in the last 4 weeks, were there any patterns of LLM token waste? For example, repeated "busy work" the agents didn't need to do, or missing tools or skills that would have made results more deterministic or less token-expensive.
```

## Improve collaboration

```text
Before we go and start building this out, use Powerplant MCP to check if anybody else on the team working on this right now. If not, go ahead.
```

```text
Steve is newly onboarding to the team this week. Based on his coding sessions so far, how might I better support his onramp? Is he hitting any undue dev or agent friction in terms of coding or access setup?
```

## Finding and handing off sessions

```text
The Betterstack alert about the log queue is assigned to Esther. Can you see in her Powerplant logs for the investigation she's doing whether we know yet if it's impacting all customers, or just free users?
```

```text
I was working on PR 38162 in a Codex session on another machine. Can you get it running for me here in Claude, recap based on my Codex session what the known issues and todos were on that work, and then recap for me the known remaining steps to getting this tested and mergable.
```

```text
I swear I had a session this week somewhere where the agent was complaining about auto-approve rejecting some of our CI commands as I was trying to get it ready to merge. Don't remember if it was in Cursor or Claude. Find it for me in Powerplant MCP, I want to fix this problem for everyone on the team.
```

## Finding your own questions

An MCP is "choose your own adventure". Different teams have different priority questions, and get different results depending on the shape of their session data. That's why we hold regular sessions with each team to check in on priorities, get feedback, and work together to find the highest-leverage ways to get the outcomes you want.

## Get help

We're rapidly improving Powerplant based on teams' feedback, so let us know how things could be better in your team's shared Slack channel with us, or email [support@usepowerplant.com](mailto:support@usepowerplant.com).