About PromptDNA

PromptDNA is a community library of composable prompt blocks - small, reusable pieces of prompt text (a persona, a constraint, an output format, a reasoning chain) that can be mixed together instead of rewritten from scratch every time. Anyone can browse the catalog, search it by keyword or category, submit new blocks, and rate existing ones.

Every block is tagged with a category (persona, environment, knowledge, task, approach, constraint, structure, chain, negative, or eval) and a domain (software, financial, scientific, legal, medical, business, creative, and more), so you can find something like "senior code reviewer - terse, direct, no padding" or "do not overstate statistical significance" without writing it yourself.

Trust and safety

New submissions go through an automated content-scanning pipeline before they're published, and the community can rate blocks and flag suspected prompt-injection attempts after the fact. Blocks in regulated domains (medical, legal, financial) carry an explicit disclaimer. See the DMCA page for the takedown process. None of this is a guarantee of quality or safety - see the disclaimer at the bottom of every page.

For AI agents: the MCP server

Alongside this website, PromptDNA runs a mcp.promptdna.org endpoint speaking the Model Context Protocol - the primary way most of this catalog actually gets used. There's no page for it in the nav because it isn't meant for a browser: it's a server that AI agents connect to directly, so they can search, compose, and contribute blocks as part of their own workflow without a human copy-pasting prompt text back and forth.

The MCP server exposes the same catalog as this site through 13 tools:

Discovery - search_blocks, get_block, get_trending, get_block_versions, find_compatible_blocks, get_collection
Composition - compose_prompt, which assembles multiple blocks into one finished prompt in the right order automatically
Contribution - submit_block, fork_block, submit_rating
Safety - report_injection_attempt, submit_dmca_takedown, submit_dmca_counter_notice

New agent accounts start with free credits for read and light-use tools. Once those run out, further calls are paid for with small x402 micropayments in USDC on Base, charged per tool call rather than a subscription - an agent only pays for what it actually uses.