Brooke Wright
Brooke Wright · @wright_mode
FREE GUIDE

Stop writing prompts from scratch

The setup that puts a 2,173-prompt library inside Claude Code and gets your agent to pick the right one for you. Plus the five prompts I'd run first.

One-line install 5 copy-paste prompts Codex version too Built by Brooke
Something went wrong. Please try again.

No spam. Unsubscribe anytime.

The MCP install line, the rule that makes your agent auto-pick, and 5 copy-paste prompts Jump to the setup → Join the Membership
Wright Mode — Free Guide

The free prompt library that plugs straight into Claude Code

prompts.chat holds 2,173 free prompts, plus skills and workflows. It also runs an MCP server — so your agent can search the whole thing and pick the right prompt itself. One line of setup, and you never copy-paste a prompt again.

One-line MCP install Works in Claude Code + Codex 5 worked examples Real prompt count, checked 07/09 Free and open source

What's inside

1 Connect it 2 Auto-pick 3 5 examples 4 The catch 5 ChatGPT + Codex

Plug the library into Claude Code

prompts.chat runs a hosted MCP server. One line in your terminal and your agent can search the whole library and pull any prompt into the chat itself — no browser tab, no copy-paste.

1

Add the server

Open your terminal, run this once. It's account-wide, so every project gets it.

claude mcp add --transport http prompts-chat https://prompts.chat/api/mcp
2

Check it actually connected

Two ways. From the terminal, claude mcp list should show prompts-chat with a green tick. Inside a Claude Code session, type /mcp — you want to see prompts-chat · connected and three tools listed.

claude mcp list

The three tools you should see

search_prompts — finds prompts by keyword, category or tag. get_prompt — pulls one by ID and fills in its blanks. save_prompt — publishes your own back to the library (this one needs a free API key; the other two just work).

3

Prove it works

Paste this into Claude Code. If it comes back with real prompt titles and authors, you're connected.

Use the prompts-chat MCP server to search for prompts about writing cold outreach emails. Show me the top three, with the author and a one-line summary of what each one actually does differently. Don't run them yet.

If you'd rather not hit their server

There's a local version that runs on your machine instead. Same tools, needs Node installed.

claude mcp add prompts-chat -- npx -y @fkadev/prompts.chat-mcp

Undo it

Changed your mind? claude mcp remove prompts-chat and it's gone. Nothing else on your machine is touched.

Make it pick the prompt for you

Installing it does not make Claude use it

This is the bit nobody mentions. Adding an MCP server makes the tools available — it doesn't make your agent reach for them. Out of the box Claude will keep writing its own prompts and never search the library once. You have to tell it when to look.

Fix it once, permanently, with a standing instruction. Open ~/.claude/CLAUDE.md (make the file if it doesn't exist) and paste this at the bottom.

## Prompt library Before you write a prompt from scratch for a repeatable job — writing, coding, image generation, research, analysis, or anything I'll want to run again — search the prompts-chat MCP server first. Rules: - Search with 2-3 different keyword angles before deciding nothing fits. One search that returns rubbish is not proof. - Show me the best 2-3 matches with the author, then tell me in one line which you'd use and why. - Adapt the one you pick to my actual context — my business, my tone, my constraints. Never paste a library prompt at me unchanged. - If nothing in the library beats what you'd write yourself, say so plainly and write your own. Do not force a bad match. - Skip the library entirely for one-off conversational questions. This is for prompts I'll reuse.

Restart the session after you save it

CLAUDE.md is read when a session starts. Save the file, close Claude Code, open it again. Then ask it for a prompt for something and watch it search first.

Don't want it standing permanently? Use it per-job instead — paste this at the top of a chat when you want the library in play and nowhere else.

For this job, search the prompts-chat MCP server before writing anything yourself. The job: [describe what you're trying to do] My context: [your business, your audience, the constraint that matters] Search a few different ways, show me the two or three best matches with authors, tell me which one you'd back and why, then adapt it to my context and run it. If the library genuinely has nothing better than what you'd write, say so and write your own instead.

Five jobs it's actually good at

One per category, plus a skills one. Paste any of these straight into Claude Code once the server's connected — each drives the library rather than asking Claude to invent something.

1

Writing — a sales page that doesn't read like AI

The library's strongest category by a mile. Roughly three quarters of what's in there is text prompts.

Search prompts-chat for copywriting and sales page prompts. Find the two that go deepest on the customer's actual objections rather than just listing features. Then combine the best of both into one prompt tailored to me: I sell [offer] to [who], at [price], and the objection I lose most often is [objection]. Show me the combined prompt first so I can read it. Then run it.
2

Coding — a proper code review before you ship

Pulls a reviewer prompt from the library and points it at real files in your project.

Search prompts-chat for code review and senior engineer prompts. Pick the one that's most specific about what to actually check — not the one with the nicest wording. Apply it to the files I changed in this project. Then give me the findings ranked by "would this break in production", with the actual fix for each, not just a description of the problem.
3

Images — a prompt that survives contact with the model

There are around 400 image prompts in there. Most of the value is in the ones with lighting and camera detail baked in.

Search prompts-chat for image prompts, filtered to type IMAGE. I want [what you're making — a product shot, a headshot, a background]. Show me the three that specify lighting, camera and composition rather than just describing a vibe. Explain in one line what each one is doing technically. Then rewrite the strongest one for my subject: [describe it]. Give me the final prompt as plain text I can paste straight into an image model.
4

Research — a competitor teardown with sources

Library prompt for the structure, your agent for the actual looking.

Search prompts-chat for research, analyst and competitor analysis prompts. Pick the one with the tightest structure — headings and sections, not vibes. Use that structure to analyse [competitor] against my business: I do [what you do] for [who]. Rules: every claim gets a source link. Where you can't verify something, write "unverified" instead of guessing. Finish with the three things they're doing that I'm not, ranked by how hard they'd be for me to copy this month.
5

Skills — turn a good prompt into a permanent one

The library carries skills as well as prompts. This is the one that compounds — a prompt you run once is a task, a skill is a thing your agent can always do.

Search prompts-chat for skills related to [the job you do over and over]. Show me what's there. Then take the best one and build me a proper Claude Code skill from it in .claude/skills/ — a folder with a SKILL.md that has name, description and clear steps. Adapt it to how I actually work: [your tools, your file structure, your standards]. When it's written, tell me the exact phrase I say to trigger it.

What's actually in there

I counted, and it's not 10,000

I checked the live prompts.chat API on 07/09/2026 before writing this page. The real total is 2,173 prompts. Plenty of people online quote 10,000+ — that number isn't coming from the library. The 170,000 figure floating around is GitHub stars on the repo, not prompts. Two thousand good ones is still the biggest free library going. It's just not ten thousand.

The breakdown, as of the day I wrote this:

📝

1,650 text prompts

Writing, coding, analysis, business, roleplay. This is the bulk of it and the strongest part.

🎨

~410 image prompts

Product shots, portraits, styles. Quality swings hard — the good ones are very good.

🧩

68 skills

Packaged, reusable jobs rather than one-off prompts. Small library, but this is where it's heading.

🎥

40 video + 4 audio

Thin. Treat these as a bonus, not a reason to install it.

So here's the honest version of what plugging this in buys you:

What it genuinely gets you

  • A starting structure for any repeatable job, in about four seconds, without leaving your terminal
  • Prompt patterns you wouldn't have thought of — the good ones are noticeably more specific than what an agent writes cold
  • An end to the notes-app prompt graveyard. You stop being the filing system
  • Free, open source, self-hostable if you'd rather own it

What it doesn't

  • Quality control. Anyone can publish — a chunk of it is thin, and votes are the only filter
  • Your context. Every prompt is generic until you adapt it. Paste one unchanged and it shows
  • Automatic use. Without the CLAUDE.md rule above, your agent ignores it completely
  • Skill and workflow search through the MCP tool. Its filter only covers text, structured, image, video and audio — skills are on the website but not in that dropdown, so ask for them by keyword instead

The way I'd actually use it

Not as a library you browse. As a second opinion your agent checks before it writes. You never open the site, you never save a prompt again, and roughly one time in three it finds an angle you'd have missed. That's worth one line of setup.

Doing this in ChatGPT or Codex instead

It's the same server. MCP is an open standard, so anything that speaks it can connect to the same URL.

1

Codex CLI

Run codex mcp add and give it the same URL, or hand-edit ~/.codex/config.toml. Then start a session and type /mcp to confirm it's connected.

2

Or paste it straight into the config

Codex keeps MCP servers in the same TOML as everything else.

[mcp_servers.prompts_chat] url = "https://prompts.chat/api/mcp"

The auto-pick rule works there too

Codex reads AGENTS.md instead of CLAUDE.md. Same file job, different name — paste the rule from the section above into ~/.codex/AGENTS.md and it behaves identically.

ChatGPT itself

Custom MCP connectors sit behind developer mode in ChatGPT's settings, and they move around a lot. If you can't find it, the honest answer is that Claude Code and Codex are where this is genuinely easy right now — and prompts.chat works fine as a plain website in the meantime.

Ready to go deeper?

Where to go from here.

Community + Training

Wright Mode Membership

Join a community of women entrepreneurs implementing AI and automation in their businesses. Live calls, templates, and ongoing support.

Learn more →
Workshop Replay

Claude Masterclass

Learn how to use Claude like a pro. From prompting fundamentals to building real workflows that save hours every week.

Watch now →
Go further

Claude Code Masterclass

Go beyond the chat interface. Build automations, process data, and create tools with Claude Code — no developer experience needed.

Explore →