BuildGist

BuildGist in Claude

Your coding agent just touched a dozen files. BuildGist tells you in about 30 seconds what it changed and what to check before you merge, then gives you a short handoff so a fresh session picks up where this one stopped, without you explaining it all again.

The BuildGist connector brings that into Claude (claude.ai, Claude Desktop and Claude Code). Give Claude a git diff, a git log -p, a Claude Code or Codex session file (.jsonl) or a Cursor chat export (.md), and ask it to use BuildGist.

Server URL

https://buildgist.com/mcp

Remote MCP server over Streamable HTTP. No account, no sign-in and no API key.

Add it to Claude

claude.ai and Claude Desktop

  1. Go to Customize → Connectors, choose + Add, then Add custom connector.
  2. Name it BuildGist, enter https://buildgist.com/mcp as the URL and choose Continue.
  3. Under Authentication, choose No sign in, then Add.
  4. In a chat, open the + menu, choose Connectors and turn on BuildGist.

On Team and Enterprise plans an owner adds it first under Organization settings → Connectors (Add, Custom, Web). Free plans allow one custom connector.

Claude Code

Claude Code

One command adds BuildGist in Claude Code. Then ask it to check what changed in your current diff and what's risky before you merge, so you don't rebuild that context by hand.

claude mcp add --transport http buildgist https://buildgist.com/mcp

Then ask it this:

Use BuildGist to explain what changed in my current git diff and what's risky before I merge. Run git diff and git diff --staged, then call explain_agent_work with that output as the text argument.

The three tools

All three only read what you send and return text. None of them writes anywhere or changes your code.

explain_agent_work

Input: text (required), the diff, git log or session, up to 400 KB; include_handoff (optional, default true).

Returns: a title and outcome, the change in plain words and for developers, what changed in short groups, and Check first: things you did not ask for, instructions the agent broke, new code nothing calls, tests that did not run. Then the Continue handoff below. It makes one AI model call; past the daily limit it returns the handoff alone and says so.

build_handoff

Input: text (required), as above.

Returns: a Continue handoff of about 400 tokens or less: your goal and the later instructions that still apply (from a session), the files changed, what the tests showed and what to check first, ready to paste into a fresh Claude Code, Codex or Cursor session. No AI model call.

get_sample

Input: none.

Returns: a built-in example: a short agent session (two requests and the agent's diff) in which the agent broke one of the developer's instructions, the explanation written for it ahead of time, and its handoff. No AI model call. Try this first to see what BuildGist returns.

Limits

Privacy

Nothing you send is stored. Keys and tokens are masked on BuildGist's server before anything reaches an AI model; for an explanation, the masked text is sent once to our AI provider (Anthropic). The handoff and the sample use no AI call. What was sent and what was returned are dropped when the answer goes back. The connector sets no cookie and never reads your other chats, memory or files. BuildGist keeps only daily counts (connects, calls of each kind, failures and the day's AI spend). Secret masking is best-effort, so leave out anything you would not paste into a chat. Details: Privacy, BuildGist in the Claude apps.

Example prompts

Want this for every run, automatically?

BuildGist for Mac explains each Claude Code, Cursor and Codex run on its own, from uncommitted work too, and offers the handoff when you switch agents. Download BuildGist for Mac or see Getting started. You can also paste one into Try it in the browser.

Support

Questions, a wrong answer or something that broke: email kayvanandre@gmail.com.