All posts
Developer tracking7 min read

Engineering report for the CEO: the monthly developer report, asked of Claude

For a CEO or a CTO who wants the same page every month: how the report is asked of Claude with the team_report prompt, which facts DeployIt returns, and what Claude composes without inventing.

The DeployIt Team

We build DeployIt, an MCP server that lets a whole company ask its code from Claude and track each developer's work.

An engineering report for the CEO does not have to be rebuilt by hand each month. With DeployIt, a person with the Leadership profile runs the team_report prompt in Claude: DeployIt returns facts drawn from merged commits, and Claude lays them out as a table.

You run the company, or its engineering team. Once a month you want one page: who delivered what, reviewed or not, tested or not, with which defects and which code score. Below is the conversation that produces it, turn by turn, with what DeployIt supplies and what Claude adds.

In your Claude · example

Give me October's engineering report, developer by developer.

Here is October for the three tracked developers. Each code score is given with the number of commits it covers.

DeveloperMerged commitsReviewedCode score
Camille R.4671%3.9 (46 commits)
Idriss B.3860%3.2 (38 commits)
Léa M.3566%4.1 (35 commits)
Drawn from your commits119 merged commits · sample data

What does an engineering report for the CEO hold?

Five facts per developer, and the same five for the team: what was merged, what a colleague reviewed, what came with tests, which defects were traced back, and the code score. Two lines make it usable: the period it covers, and the time the figures were last computed. Everything else is commentary.

Step 1: who asks, and with which profile

The report belongs to the Leadership profile. Everyone who signs up through your company link starts with the Everyone profile, which has no statistics; an administrator gives Leadership to the people who should read each developer's work. Leadership adds the statistics of the team and of each developer to the questions answered without code.

Statistics need the trial or a plan. The free plan has none. During the trial, tracking starts as soon as your repositories are connected, with 6 months of history, so the first report you ask for can cover a month that is already closed.

Step 2: run team_report with a period unit and a period

You run the prompt: team_report, unit month, period 2026-10.

team_report is one of the three prompts DeployIt offers. In the Model Context Protocol, a prompt is a template that a server exposes and that the user selects explicitly, with arguments to fill in (MCP specification, Prompts, checked on 17 November 2026). This one exists so that the table has the same shape every month.

It takes two arguments, both optional:

  • period_unit: month, quarter, semester (the half-year) or year;
  • period: which one. Example, for October 2026 read by the month: unit month, period 2026-10.

Give both. A report that names its period can be set next to the following one.

Where to find it, as checked on 17 November 2026: the MCP documentation shows a connected server's prompts in the menu at the bottom left of Claude's message box, under Connectors (connect to remote MCP servers). In Claude Code, the documentation gives the form /mcp__servername__promptname, with the arguments after it, separated by spaces. With a server added under the name deployit, that is /mcp__deployit__team_report month 2026-10.

You can also ask in your own words, as in the example above. The prompt is for the report you want identical from month to month.

Step 3: Claude queries, DeployIt returns facts

Claude calls stats_schema, then stats_query. DeployIt answers with rows.

Claude first calls stats_schema: the datasets, the definition of each measure, and when they were last computed. Then it calls stats_query, a structured query: a dataset, a period, filters, groupings, measures, a limit. Free-form SQL is not accepted.

For a monthly report, the rows come from the summaries by developer and by team (person_period, team_period). They carry:

  • merged commits: only work that reached each repository's landing branch, counted when it lands;
  • reviewed by a colleague: neither the author nor a bot;
  • tested, or not;
  • defects, “to confirm” or “confirmed”;
  • the code score, from 1 to 5, with the number of commits it covers;
  • the time the summary was last computed.

These facts are written once per merged commit and updated every night. Every row links back to its commits. With the Leadership profile, a query returns at most 500 rows and answers within 10 seconds.

Step 4: Claude composes the page

Claude replies with the page: the period, the table, the cautions.

The layout is Claude's work. It orders the rows, spells out the period and writes the sentence above the table. The same month in full, as an example with sample data:

DeveloperMerged commitsReviewedTestedDefectsCode score
Camille R.4671%52%1 (to confirm)3.9 (46 commits)
Idriss B.3860%31%2 (1 to confirm)3.2 (38 commits)
Léa M.3566%40%04.1 (35 commits)

The conversation does not stop at the table. Two follow-ups a CEO asks next:

  • “Which commits are behind Idriss's two defects?” Claude queries the defects: for each one, the commit it came from, its fix, who fixed it, and whether it is “confirmed” or “to confirm”.
  • “And over the quarter?” Run team_report again with the unit quarter.

For one person, the developer_profile prompt returns that developer's record over a period. It is the page to bring to a conversation with them: see engineering 1:1s anchored in delivery.

Step 5: what Claude must not invent

You check: “Which rows does that sentence rest on?”

Three things are not in the rows. Claude has no source for them.

A missing figure. If a value did not come back, it is missing. Ask for a dash, not an estimate.

A reason. DeployIt measures commits, pull requests and reviews. It does not see a screen, messages or the time spent on a task. “October was slow because of the migration” is a guess, unless you supplied that fact yourself. The only context DeployIt holds is what the company declares: absences and working time, so that a quiet month is not read as a drop.

A verdict on a person. A score set by a model is an indication, never on its own the reason for a decision. A defect counted by the machine stays “to confirm” until an administrator or an automated investigation confirms it; only an administrator dismisses it, and the human decision prevails.

The check takes one question. For any sentence in the report, ask Claude which rows it rests on. A model can be wrong; a figure that leads back to its commits can be verified.

How do you read the code score on the report?

Always with its number of commits. Each commit is read by a language model, which scores it from 1 to 5 on correctness, maintainability, robustness, tests and security. The figure on the report is an average of those readings.

Example: 4.6 over 3 commits says less than 3.9 over 46. On a thin month, read the quarter; the score is available by month, quarter, half-year or year. Which per-developer figures hold up, and what they leave out, is covered in developer performance metrics from merged commits.

Three cautions that belong on the page

  • Defects arrive late. They show up after the fact: the last 60 days are not mature yet, so October's count can still rise.
  • A closed month can still move. A commit that landed late, two identities merged, a defect dismissed: hence the time of last computation on every summary.
  • Who is counted. Bots are excluded, detected coding agents are counted separately, and an undeclared agent counts as human work.

The report is not sent by email, and there is no reporting screen to open: you ask for it when you want it. The datasets and the reading rules are listed in the statistics section of the docs.

What DeployIt does here

DeployIt is an MCP server. Your company gets one link, https://mcp.deployit.ai/<company>/<code>; each person adds it to Claude (on the web, on the desktop or in Claude Code) and signs up with a name and an email, which is not verified; their questions are logged under that name. Everyone starts with the Everyone profile: questions about the product, answered from the code; a filter removes the code, the file paths and the names of team members it recognises from the answers, without proving their absence, and a leak test tells you what still gets through. An administrator gives the Developers profile (excerpts) or the Leadership profile (developer statistics) to whoever needs it. Repositories are connected from GitHub or GitLab; DeployIt makes no commit, branch or pull request, and only adds one webhook per repository.

DeployIt tracks each developer, commit by commit: the company informs the people it tracks before connecting its repositories, or turns statistics off. The rest of the product is described in DeployIt: ask your code and follow each developer's work from Claude.

Next step: open the trial, connect a repository, give the Leadership profile to whoever reads the report, and run team_report on last month. Adding your company link to Claude takes the three steps of the docs.

Price. Free plan: 15 questions a week for the whole company. A 30-day trial, no card. Plans from €290 a month excluding tax for up to 25 tracked developers: see pricing.

Ask your code from Claude

30-day trial, no card. Or try the demo first: add https://mcp.deployit.ai/demo to your Claude connectors.

Connect my code

Frequently asked questions

What should an engineering report for a CEO include?

For the team and for each tracked developer, over a named period: the commits merged, the share reviewed by a colleague, the share tested, the defects “to confirm” and “confirmed”, and the code score with the number of commits it covers. Add the time the figures were last computed. With DeployIt, these facts come from merged commits and Claude lays them out.

How do I get a monthly engineering report from Claude?

Connect your repositories to DeployIt, add your company link to Claude, and have an administrator give you the Leadership profile. Then run the team_report prompt with a period unit and a period, or ask in plain words. Claude calls stats_schema, then stats_query, and composes the table. Statistics need the trial or a plan: the free plan has none.

Can Claude make up numbers in the report?

A model can be wrong, which is why every figure should come from a row DeployIt returned. Each row links back to its commits, so you can ask Claude which commits stand behind a number. A reason, a forecast or a judgement about a person is not in those rows: read it as wording, not as a fact.

Is the engineering report sent by email every month?

DeployIt does not send the report by email. A person with the Leadership profile asks for the report in Claude when they want it, and the team_report prompt keeps the same table from one month to the next. A warehouse or a BI tool can also read the statistics with a Data tool key.

Why did last month's figures change when I asked again?

A closed period can still move: a commit that landed late, two identities merged, a defect dismissed. Defects also show up after the fact, so the last 60 days are not mature yet. Every summary carries the time it was last computed: keep that line on the page.

Continue reading

Connect my code