Session 1: What's Possible with Claude/ChatGPT + Clockwork?

The session opens with the basics: what an MCP server actually is. It's described as a secure bridge between the AI assistant you already use (Claude / ChatGPT / CoPilot, etc.) and the live data sitting inside Clockwork. Without it, your AI assistant only knows what you type into the chat or copy and paste from Clockwork. With it, your AI assistant can look at your real data and Clockwork records.

Think of an MCP Server like a GPS. The roads already existed, GPS just removes the guesswork of getting where you're going. The MCP server does the same for your AI assistant, and for you: it takes over the administrative work of moving information between systems, and provides structured directions around how to use data points, records, etc.

In this live walkthrough we cover the setup. Before connecting, you need three things: you're logged into Clockwork, the MCP Server feature is enabled on your account (visible at the bottom of your Clockwork user profile page), and you have a paid AI assistant account (Claude / ChatGPT / CoPilot, etc.). From there, prompt or ask your AI Assistant directly how to connect to Clockwork's MCP Server. 

Once connected, your AI assistant shows a list of available tools, currently 23, covering things like company records, people, project fees, locations, and teams. Setting permissions to "always allow" gives the smoothest experience, though "needs approval" and custom scoping are also available for anyone who wants tighter controls.

A recurring theme is the shift in mindset from asking for information to asking for outcomes. Instead of "find me engineers in New York," think "look across my active searches and tell me what needs my attention today," or "draft this week's client update for this project." This is a language shift as much as a technical one: the more clearly you instruct your AI assistant, the better and more consistent the results.

Two live examples make this concrete. First, a status report: a simple prompt referencing a project pulls in visible clients, flags do-not-contact candidates, and builds out a formatted report, while separating anything not marked client-visible into a clearly labeled internal-only section. This client-safety behavior is built into the MCP server by design; it uses Clockwork's existing visibility and stoplight settings, so anything not meant for a client ever leaks into a client-facing document.

Second, a candidate resume is generated from a template with a short prompt (essentially "generate a new resume for this person using this template"), reviewed, adjusted for formatting, and regenerated for a different candidate, with the AI assistant remembering the prior corrections.

Templates come up repeatedly as the highest-leverage input. Building out your firm's branding, layout, and fields once and handing that to your AI assistant means it reuses that structure every time, whether for a status report or a candidate profile, pulling live Clockwork data into the format you've already defined. 

On consistency and complexity, the guidance is to keep prompts short and conversational rather than long and dense. Anything beyond two or three instructions in a single message tends to produce weaker results; breaking a request into smaller, sequential asks works better than one long, complex prompt. Guardrails matter here too: giving your AI assistant a target number of results, a timeframe, or a specific scope (for example, "top 10 candidates from the last 60 days who've had recent activity") produces a tighter, more useful response than an open-ended ask that could return hundreds of records.

Memory and skills are different things: memory is what an AI assistant retains within an ongoing chat or project, letting you keep building on prior context without restating it. While skills are a more formal, reusable set of instructions your AI assistant follows every time (a Claude-specific term, though every AI assistant has some version of the concept). 

Attendee feedback compares Claude and ChatGPT directly. Claude is the clear favorite for documents, spreadsheets, PowerPoint templates, and branded content, with several attendees specifically praising its handling of PowerPoint files. Image generation is the notable exception, with multiple attendees calling Claude weak on custom images and preferring ChatGPT or Gemini for that use case. One workflow shared in the chat: use Claude for analytical and data work, then hand that output to ChatGPT to refine tone and match personal writing style. Attendees also confirm that giving an AI assistant an existing template or document to match tends to work better than asking it to design a template from scratch.

The session closes with a look ahead. A full AI prompt playbook for Clockwork users, with example prompts organized by use case and adaptation suggestions for each, goes out to all attendees.

Workshop 2 goes deeper for partners and owners, and Workshop 3, "Data to Deliverables," focuses on custom reporting.