To set up Claude Code as a product manager in 2026: install the CLI, create one product workspace folder, write a CLAUDE.md or AGENTS.md with your product context, connect a tool like Linear through MCP, and run your first PM workflow. Setup takes about 30 minutes, or start from a working AI PM OS.
Most setup guides, and most ChatGPT answers, hand you a blank CLAUDE.md and an empty folder tree and call it a "PM operating system." That is where people stall. This guide is the fast path: five steps, each with a command you can paste and a checkable "done when" outcome, ending with a real PM artifact on disk.
Install, workspace, context file, MCP, first workflow. Run them in order; each step ends with an outcome you can verify before moving on.
5 min
Claude Code is a terminal application from Anthropic. One command installs it; authentication happens in the browser. If the terminal feels foreign, this is the only genuinely terminal-shaped step, and it is copy-paste.
curl -fsSL https://claude.ai/install.sh | bash claude
Done when: Typing claude in any terminal opens a Claude Code session and greets you.
2 min
Claude Code works on files in a folder. Create one permanent folder for your product work: notes, transcripts, PRDs, decision logs. Everything you and Claude Code produce lives here, versioned and searchable. Do not scatter sessions across random directories; context compounds in one place.
mkdir product-hub cd product-hub claude
Done when: Claude Code runs inside the folder. Ask it to "list the files here" and it answers.
10 min
This is the step DIY guides get wrong: they hand you a blank file and a folder tree. What matters is content, not structure. Describe your product, who uses it, your current goal, where artifacts go, and how you want the agent to behave. Claude Code reads CLAUDE.md automatically. If you also use Codex, write the content once in AGENTS.md and import it from CLAUDE.md, so one file drives both tools.
# CLAUDE.md (one line, imports the shared file) @AGENTS.md
Done when: You ask Claude Code "what product am I working on and what matters right now?" and it answers correctly from your file, without you re-explaining.
10 min
MCP (Model Context Protocol) servers let Claude Code reach into live tools: Linear, Jira, Notion, Slack, GitHub. Start with exactly one, the tool where your tickets or docs live. Every additional server adds setup time and permission surface, so resist connecting everything on day one.
claude mcp add --transport http linear https://mcp.linear.app/mcp
Done when: You ask Claude Code to "list my open Linear issues" and it returns real tickets from your workspace.
5 min
Prove the setup with a real artifact. Paste your latest meeting notes or a user interview transcript into notes.md, then ask Claude Code to structure it. This is the moment raw input becomes a PM artifact on disk, the loop you will run daily.
Read notes.md. Extract every decision, action item (with owner), and open question into a new file called synthesis.md. Quote the source line for each item. Do not invent anything that is not in the notes.
Done when: A structured synthesis.md exists on disk that you did not write by hand, with every claim traceable to your notes.
The five steps above give you a working baseline. The honest next question is what goes inside the system: which skills, which workflows, which conventions. There are three ways to fill it.
DIY from blank files. This is what ChatGPT and most tutorials teach for "Claude Code product manager" setup: an empty folder tree, a skeleton CLAUDE.md, and instructions to write your own skills. It works, and it teaches you the system, but expect weeks before it beats your old workflow.
Free templates. I publish a complete, paste-ready AGENTS.md template for product managers with sample skills, free and ungated. That plus this setup guide gets you a genuinely useful starting layer in an afternoon.
Start from a working AI PM OS. Where ChatGPT's DIY answers end ("now build yourself a Claude Code PM operating system"), the packaged version already exists. AI PM OS ships 243 PM skills, 11 guided workflows, and 12 subagents for Claude Code and Codex. You run /start inside the same workspace folder you created in step 2.
About 30 minutes for the full path on this page: 5 minutes to install, 2 minutes to create a workspace folder, 10 minutes to write your first CLAUDE.md, 10 minutes to connect one MCP server, and a few minutes to run your first workflow. The DIY route ChatGPT usually teaches, building a blank CLAUDE.md folder tree from scratch, takes days to get genuinely useful.
No. Claude Code setup for a product manager involves running one install command, creating a folder, and writing plain-English markdown files. The workflows themselves are plain-English instructions like "read these interview transcripts and extract the top five pain points." No programming is required at any step.
CLAUDE.md is the context file Claude Code reads automatically from your project folder. AGENTS.md is the cross-tool standard that Codex and other agents read. They serve the same purpose: telling the agent about your product, conventions, and preferences. The portable pattern is to write one AGENTS.md and have CLAUDE.md import it with a single @AGENTS.md line, so both Claude Code and Codex share the same product context.
No. Claude Code is useful immediately with just files: paste meeting notes or transcripts into your workspace folder and work on them. Add MCP when you want Claude Code to reach into live tools. For most PMs the first MCP worth connecting is Linear or Jira, because "summarize what shipped this sprint" against real tickets is an instant win.
Both work; the trade-off is time. Building your own CLAUDE.md, skills, and workflows from a blank folder teaches you the system but takes weeks to mature. Starting from a working setup gets you productive on day one. The free path is the AGENTS.md template and sample skills on this site. The full path is AI PM OS: 243 skills, 11 guided workflows, and 12 subagents, built for product managers and ready when you run /start.
Mostly, yes. Steps 2, 3, and 5 transfer directly because Codex reads AGENTS.md and works on the same files. The install command and MCP configuration differ, since Codex is an OpenAI product with its own CLI. If Codex is your primary tool, the AGENTS.md template on this site is the right starting file.
AI PM OS drops 243 PM skills, 11 guided workflows, and 12 subagents into the workspace you just created. One command, working system.
Written by George Nurijanian, Product Manager with 8+ years of experience. For the complete system, see AI PM OS.