All posts
Product

DeployIt: ask your code and follow each developer's work from Claude

One link for your company's Claude: everyone asks how the product behaves and gets answers from the code without seeing it (only the Developers profile sees excerpts), and leadership reads statistics on each developer's work.

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.

DeployIt is an MCP server that connects a company's repositories to its Claude: everyone asks their questions about the product there, leadership reads statistics on each developer's work, and the source code stays closed to everyone but the developers you choose.

This page describes what DeployIt does, as it works today: how people connect, what each profile sees, the two things you can ask, where your code goes, and what it costs.

Your company gets one address, of the form https://mcp.deployit.ai/<company>/<code>. The first part identifies your company; the second is a code that opens sign-up.

In Claude, open Settings, then Connectors, and add a custom connector with that link. On a Claude Team or Enterprise plan, an owner adds it once for the whole organization, and each member then clicks "Connect". Claude opens a DeployIt page: the person gives a name and an email, and that is the sign-up. The email is not verified, because the code is the door. Every new sign-up starts with the Everyone profile.

The same link works in Claude Code (claude mcp add --transport http deployit <link>). Claude on the web and desktop, and local clients such as Claude Code, can sign up; Cursor and Windsurf cannot yet. Keep the link out of any file you commit: it carries your company's code.

If the link leaves the company, an administrator changes the code. Nobody can sign up through the old one, and people already signed up keep their access. Removing a person cuts their access without touching anyone else's.

If your Claude organization manages sign-in through its identity provider, automatic connection is available in beta, on request: write to us. An administrator then declares it in DeployIt, and people are signed up without typing anything. Their address counts as attested when the issuer is Claude or when your identity provider marks it verified; otherwise it stays declared. In the automatic connection settings of your Claude organization, always fill "Tenant ID" with your company's identifier in DeployIt, the part of the link after mcp.deployit.ai/; once automatic connection is on, that identifier no longer changes.

Four profiles

A profile is attached to a person, never to an address. An administrator raises the profile of a sign-up they can see; signing up again with the same address gives Everyone back, never more.

ProfileCodeDevelopers' namesStatistics
EveryoneNo line, no file pathRemoved, when DeployIt knows themNone
DevelopersExcerpts and pathsShownNone
LeadershipNo line, no file path: commit subjects, pull request titles, module names, links to commits and pull requestsShownTeam and individuals
Data toolNo question about the code; statistics carry the detail (modules, paths)ShownTeam and individuals, in bulk

An output filter removes what it recognizes: code blocks and fragments, file paths, secrets and personal email addresses. It does not prove their absence. Administrators can also list, profile by profile, topics the assistant must not discuss, and every request is logged: who asked what, and what was answered.

Asking how the product behaves

The first use is the question support, product and sales teams ask all day: what does the product actually do? Take this question, as an example: "Why does a customer see 'invalid IBAN' at sign-up?" Claude calls the ask_code tool, which investigates in the code before answering. The first time, an investigation takes 20 to 60 seconds; once the answer is known, it comes back at once for the same question asked in the same words and without context, as long as the code has not changed.

Every answer carries a status, and the status says whether you can repeat it:

  • Verified: confirmed in the code.
  • Backend only: the product can do it, but no screen leads there.
  • Not found: looked for without being found; most likely absent from the product, which is not proof.
  • No conclusion: the investigation did not settle it. Do not repeat it as a fact; ask again with the in-depth option if your profile allows it, or ask someone with the Developers profile.

The second tool, recent_changes, answers "what changed this week?" by listing what landed over a period (1 to 90 days, 14 by default), grouped by pull request, or by day, module and type, without anyone's name. To understand one change, ask ask_code, which reads the current code and the commits of the last 30 days at most: their messages and dates for every profile, their diffs as well for the Developers profile only. The what_changed prompt asks it the same way every time.

Following each developer

The second use is for leadership. DeployIt turns each merged commit into a fact, attributed to its author: shipped, reviewed by a colleague or not (not its author, not a bot), tested or not, fixed since or not, and what a model's reading of the code says about it.

A defect is counted when DeployIt finds the commit that a fix came to repair. As long as only the machine has counted it, it is "to confirm"; an administrator, or an investigation that rereads the code, can mark it "confirmed"; only an administrator can dismiss it, and their decision prevails. Defects take time to surface, so the last 60 days are not yet mature, and the statistics say so.

The code score is a language model's reading of each commit, from 1 to 5, on five criteria: correctness, maintainability, robustness, tests and security. It is read over a month, a quarter, a half-year or a year, and it is always given with the number of commits it covers.

Leadership asks in plain language: "Give me October, developer by developer", or "Which modules depend on a single person?" Claude calls stats_schema to learn what exists, then stats_query, a structured request (dataset, period, filters, groupings, measures, limit), and composes the table you asked for. Two prompts keep a recurring report identical from one month to the next: team_report for the team over a period, developer_profile for one developer. Claude composes the table; DeployIt supplies the facts, each of them traceable to its commits.

A warehouse or a BI tool does not sign up. It presents a data-tool key, created by an administrator and shown once, in a request header. That key reads statistics in bulk and asks nothing about the code.

What it does not do

DeployIt has no dashboard and sends no report by email: the tables are the ones you ask Claude for. It accepts no free-form SQL. It writes nothing to your repositories but one webhook per repository: no commit, no branch, no pull request comes from it.

A leak test attacks each profile that should not see code and tells you what came out. It runs on demand, and automatically after a settings change that affects the answers (at most ten a day). It warns you; it blocks nothing, and no filter is infallible.

A model can be wrong. That is why every answer about the code carries its status, and why a score is read with its number of commits.

Where your code goes

Repositories are connected from GitHub or GitLab, only the ones you choose; DeployIt makes no commit, branch or pull request there, and only adds one webhook per repository. A working copy of the code is kept while a repository is connected, hosted by DigitalOcean in London (UK); removing a repository from DeployIt erases its copy.

At indexing, each source file is sent to the model to be summarised; then the passages needed for an answer, and the diffs of the commits that are scored, are sent to OpenAI's models, or to the provider whose key you bring. DeployIt does not use your code to train models; with your own key, your provider's terms apply.

When a subscription ends, the company moves to the free plan: nothing is erased at that moment, but developer tracking stops and the statistics close. Deleting the company (possible once the last paid period has ended) cuts all access at once, then a queued job erases its data, usually within the following hours, and retries if it fails; the daily database backups, which DigitalOcean keeps for seven days, hold it until they rotate. The full path of your code, profile by profile, is on the security page.

Telling the people you track

DeployIt does track each developer, during the trial or on a plan, from the moment the repositories are connected, with 6 months of history; the company informs the people it tracks before connecting them, or turns statistics off, and a model's score is a clue, never on its own the reason for a decision about someone. The terms of service set out the framework for this use.

You can export everything DeployIt knows about a tracked person, and erase it. What a tracked developer can expect is described on the privacy page.

Price

The free plan gives your whole company 15 questions about the code a week, Monday to Sunday; an answer that is already known does not count. It has no developer statistics.

The 30-day trial needs no card, and opens once per account, with its first company. Everything is open, for up to 25 tracked developers. Without a plan at the end of the trial, your company moves to the free plan.

Three plans, priced by the number of tracked developers and excluding tax: Team, €290 a month for up to 25; Business, €590 for up to 60; Group, €1,190 for up to 200. Paying yearly, you pay ten months for twelve. Beyond 200 tracked developers, the price is quoted. Repositories, profiles and everyone who asks are included; questions are covered up to a monthly fair-use limit per plan, shown in your workspace, and each Everyone sign-up is capped at 40 a month by default, which an administrator can change.

The limit holds at all times: DeployIt tracks active people automatically while the plan has room, and you cannot track more developers than your plan covers, so there is never an overage. A newcomer beyond the limit is not tracked, and administrators are told. Coding agents and bots never count. Details are in the pricing section.

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

Who gets which profile?

Everyone who signs up through your company link gets Everyone. An administrator gives Developers or Leadership to whoever they choose. Signing up again with the same address never gives a broader profile.

Is developer tracking individual?

Yes, that is what it is for. Each tracked developer has a record: what they shipped, reviewed, tested and fixed, and the score of their code. During the trial or on a plan, tracking starts as soon as your repositories are connected, with 6 months of history: inform your teams beforehand, or turn statistics off. The terms of service set the rules, and we give you what you need to inform people, including an export of a person's data.

What is the code score?

A language model reads each commit and scores it from 1 to 5 on five criteria: correctness, maintainability, robustness, tests and security. It is read over a month, a quarter, a half-year or a year, always with the number of commits it covers. It is a clue, not a verdict.

What if we stop?

When your subscription ends, your company moves to the free plan: nothing is erased at that moment, but developer tracking stops and the statistics close; the data is kept for its retention period. Export it beforehand, while your plan or the trial runs, with a Data tool key, and each person's record from Developers. Removing a repository from DeployIt deletes its working copy and, if you tick the box, its statistics. Deleting the company (possible once the last paid period has ended) cuts all access at once, then a queued job erases its data, usually within the following hours, and retries if it fails; the daily database backups, which DigitalOcean keeps for seven days, hold it until they rotate.

Continue reading