MCP App Store
by KeenEthics
DocketDrafter icon

DocketDrafter MCP Server for ChatGPT

By DocketDrafter

Overview

DocketDrafter gives attorneys a private, firm-specific document workspace in ChatGPT. It helps the user's agent reuse prior work, draft and revise legal documents in the firm's preferred court formatting, preserve version history, produce clean Word files and redlines, manage source documents, and track citation-checking work.

How to connect

  1. ChatGPT: Open the DocketDrafter listing in the ChatGPT app directory and choose Connect. ChatGPT listing DocketDrafter exposes 5 tools here.

Tools

add_files

ChatGPT
Add a .docx or .pdf the lawyer supplied to the firm's workspace. Give a path you can read from this session for each file; returns one command per file. Run each with your shell to move the bytes (the file content never passes through you), and each prints back an oid and size on success. Then, inside execute, record each file in the repository with docket add-file <path> <oid> <size> and commit, so it becomes a retrievable part of the case. Recording a file stores a REFERENCE, not its bytes, so it is still not readable in your working copy afterwards: to open one (read a PDF, or inspect a Word file's numbering and styles), list it in needs on the execute call that reads it — the oid for a file you just uploaded, or its repository path once recorded — and only then are its bytes there to read. .docx and .pdf are case content and are recorded this way. A .png or .jpg is also accepted, but it is a firm image (a signature, a letterhead mark) rather than case content: record it with docket add-asset <name> <oid> instead, which stores it in the firm's library for every document that references it. A .md is accepted only to hand a document to DocketDrafter, cited by its oid in send_feedback; it is never recorded and never read back, so firm knowledge in a .md belongs in the repository, written through execute. Nothing else is accepted. The commands are for you to run; the lawyer cannot run commands, so they are not something to show them. A command failing because your session cannot reach our host is the one case to involve them: tell them the file is ready but the transfer is blocked, and give them the domain to allow (where to allow it differs between assistants, so name the domain and let them find the setting).

execute

ChatGPT
The primary tool for doing work in the firm's repository. Run a shell script in a sandboxed working copy of the firm's files to read their prior work, draft and revise documents, and organize the repo, composing ordinary Unix tools, git, and the firm's docket commands in a single script (code mode: do as much as you need in one run). Call get_guide first; it explains the repository, the document format, and the commands available here. Work is saved only when you commit it, and the sandbox has no network access.

get_files

ChatGPT
Get files to hand to the user. For a document you authored (a .html source), renders it to a finished Word .docx. For a .cite-check.json review record, renders an Excel .xlsx status workbook. For a stored file the firm holds (a .pdf or .docx a lawyer brought in earlier, named at its real path), returns its original bytes unchanged. Either way you get a download command rather than the bytes: run it with Bash to fetch the files into this session, then present the files themselves. The user cannot run commands, so the command is not something to show them. Keep each rendered filename exactly as returned, it is this firm's filing convention. Documents render from the latest commit, so commit your changes first. A document delivers either clean or as a redline of what changed; see before. Its review comments are left out unless you ask for them; see comments. The command failing because your session cannot reach our host is the one case to involve the user: tell them the documents are ready but the transfer is blocked, and give them the domain to allow (where to allow it differs between assistants, so name the domain and let them find the setting). See get_guide for the repository layout.

get_guide

ChatGPT
Learn how to work in this firm's repository: its layout, the document format, and the commands available here. Call this once when beginning work in a conversation. Reuse the returned guide and guide_version; do not call this again in the same conversation unless execute reports that the guide version is stale.

send_feedback

ChatGPT
Relay a problem to DocketDrafter to help improve the product: a wrong or confusing result, a drafting scenario the product handles poorly, a capability the user needs that the product cannot do, or something that behaved differently than the guide or a command led you to expect. Solve obstacles yourself as you work; use this when something behaved unexpectedly, OR when the user asked for something the product cannot do or was dissatisfied with a result (report that even though you already knew the limit) — reflecting at the end of your turn and batching everything worth reporting into one call (report it even if you worked around it). Describe the problem with enough detail to be actionable — quote the wrong result or the relevant passage; do not dump an entire document or the whole conversation. Do not use it for the normal draft-and-revise loop or a user simply changing their mind.

Capabilities

Writes

Similar Apps

Listed on MCP App Store

Listed on MCP App Store

Add this badge to your site or README. It links to this page.

HTML

<a href="https://mcp-app-store.keenethics-labs.com/app/docketdrafter"><img src="https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg" alt="Listed on MCP App Store" width="220" height="44"></a>

Markdown

[![Listed on MCP App Store](https://mcp-app-store.keenethics-labs.com/badges/listed-on-mcp-app-store.svg)](https://mcp-app-store.keenethics-labs.com/app/docketdrafter)