Prompt Buddy

official

Prompt Buddy MCP exposes a public, searchable catalog of reusable AI skills.

View on prompt-buddy.io

What can you do with Prompt Buddy MCP?

  • Index local instruction skills — Run pb detect --yes to scan a repository and refresh managed skill metadata for all local instruction skills.
  • Create reusable skill packages — Use pb skills create to scaffold a local skill with a name and description for a repeatable workflow.
  • Verify skill integrity in CI — Run pb skills verify --strict to check local skill files, manifests, and checksums, failing on untrusted changes.
  • Retrieve workspace prompts via MCP — Call get_workspace_prompts with a bearer token to load authenticated workspace prompt assets into an AI client.
  • Connect an MCP client to Prompt Buddy — Point any streamable HTTP MCP client at the remote endpoint to search and load skills live.

Documentation

Prompt Buddy

DocsScore instructions

Developer docs

Manage local instructions or connect workspace prompts through MCP

Use the Prompt Buddy CLI to create, index, and verify local instruction skills. Use the remote MCP server to give AI clients authenticated access to workspace prompt assets.

Install CLIConnect MCPSkill design guide

Recommended path

Start with the CLI, then add MCP where AI clients need live retrieval.

pb detect --yes
pb skills create code-review --description "Review changes with prioritized findings"
pb skills verify --strict

CLI commands create local project state, lock installed skills, and make skill validation usable in scripts and CI.

CLI

Download and install Prompt Buddy CLI

The CLI is the fastest way to bring Prompt Buddy into a code repository. Create or detect local instruction skills, track managed state in lockfiles, and verify skills in CI.

Works as `prompt-buddy` for scripts and `pb` for daily use.

Indexes repository-local skills and their supporting resources.

Writes local Prompt Buddy state only when commands install or detect skills.

Local development

cargo install --path ./prompt-buddy-cli/crates/prompt-buddy-cli --force

Use this while Prompt Buddy CLI is developed from the monorepo.

Crates.io release

cargo install prompt-buddy-cli

Use this after the CLI package is published.

Binary releases

prompt-buddy_<version>_<platform>

Prebuilt Windows, macOS, and Linux binaries should be distributed from this docs page after cargo-dist release builds are available.

pb detect --yes

Index repository-local instruction skills and refresh their managed skill metadata.

pb skills create --description

Create a local skill package for a reusable procedure or tuned instruction workflow.

pb skills verify --strict

Check local skill files, checksums, manifests, and CI readiness.

pb clean --all --yes

Remove Prompt Buddy project state and managed agent bridges when decommissioning an integration.

MCP

Connect AI clients to Prompt Buddy skills

MCP is the live retrieval path. Point any streamable HTTP MCP client at the endpoint, then let the client search skills, inspect full instructions, or load authenticated workspace prompt assets.

Check MCP status

Remote endpoint

Public skill discovery works without authentication. Send an Authorization bearer token only for workspace-scoped workflow tools.

https://mcp.prompt-buddy.io/mcp

get_workspace_prompts

Read authenticated workspace prompt assets. This tool requires a bearer token.

get_workspace_workflows

Backward-compatible alias for clients that still call workspace prompt assets workflows.

Client setup

Add Prompt Buddy as a remote MCP server in your client. If your client asks for transport, choose streamable HTTP.

{
  "mcpServers": {
    "prompt-buddy": {
      "url": "https://mcp.prompt-buddy.io/mcp"
    }
  }
}

Tune before reuse

Use the instruction improver to score rough workflow instructions before packaging them as a local skill.

Keep docs crawlable

CLI and MCP instructions are both published as static sections so users, search engines, and AI answer engines can extract the setup path.

Verify in CI

`pb skills verify --strict` fails when local skills are missing, changed, or no longer trusted enough for production workflows.