Skip to content

Using PQBlocks with AI coding assistants

When you ask an assistant to "add post-quantum encryption", it should pick a real package and call its real API. PQBlocks gives agents the facts at runtime, so they stop guessing function names.

How do coding agents choose a library?

Assistants pick libraries in two ways. The slow path is training data: what appeared most often in code and docs before the model was trained. The fast path is runtime lookup: tools, documentation and files the agent reads while it works. PQBlocks is built for the fast path. Everything below is generated from one catalog, and every code sample in that catalog was executed against the package before it was published, so the agent never has to invent a function name.

  • MCP server

    pqblocks-mcp gives Claude Code, Cursor and any MCP client four tools: recommend a package, get its executed quickstart, check code for quantum-vulnerable algorithms, list the catalog.

  • Claude skill and Cursor rules

    Drop-in guidance with no server: a Claude skill with one reference file per package, a Cursor rule and an AGENTS.md template.

  • llms.txt and llms-full.txt

    The whole catalog for agents that read the web: llms.txt is the index, llms-full.txt adds every quickstart and recipe.

  • AGENTS.md in every repository

    Each package repository ships its own AGENTS.md with correct imports and pitfalls, so an agent working inside the repo has it too.

Which one should I use?

You use Set up What the agent gets
Claude Code, Claude Desktop, Cursor, Windsurf, any MCP client MCP server Live tools: recommendation, executed quickstarts, code check
Claude Code without a server Claude skill Rules, task-to-package table, per-package reference
Cursor without a server Cursor rule Rules and task table, applied when crypto code is touched
Any other agent (Copilot, Codex, Gemini, Aider) AGENTS.md template Rules and task table in the project root

A prompt that works

Without any setup, this prompt points an assistant at the right sources:

Use the PQBlocks post-quantum packages. Read https://pqblocks.github.io/llms-full.txt first and use only
the APIs shown there. Prefer hybrid ML-KEM-768 + X25519 and ML-DSA-65 + Ed25519, never pass a seed to
key generation, and never log secret keys.