Developers

MCP Server

The Model Context Protocol (MCP) exposes your ParetoStudio library to any MCP-compatible AI client. Connect once and Codex, Claude Desktop, Cursor, Claude Code, and others can read, search, create, and manage items in your library from natural language. Skill installation is client-aware: Codex uses .agents/skills/, while Claude uses the .claude/ layout.

Current ParetoStudio Library with one saved production prompt, search, filters, and workspace controls — Everything in this source of truth is addressable from the IDE. 21 tools, one API key, zero copy-paste.

Prefer a guided setup?

Use the setup wizard at /setup/mcp. It picks your client, creates an API key, and verifies the connection in four steps. Continue reading if you want the manual walkthrough.

Prerequisites

ParetoStudio account

Any plan including Free. Free gets 50 requests/day; Pro unlocks 500/day and unlimited saves.

API key

Generate one in Settings → Developer. Starts with psk_.

Node.js 18+

npx paretostudio-mcp@latest fetches the latest bundle. No global install required. View on npm →

Setup

Paste the ParetoStudio MCP block into your client's config, fill in your PARETOSTUDIO_API_KEY, and restart the client. Pick the tab that matches where you work.

Settings - Developer panel showing MCP Server Configuration block ready to copy — Developer tab → copy the config → paste into your client.

Config path: ~/.codex/config.toml

Add this block to your existing config without replacing unrelated settings. Start a new session after saving. Install native project skills with client: "codex".

~/.codex/config.tomltoml

[mcp_servers.paretostudio]
command = "npx"
args = ["-y", "paretostudio-mcp@latest"]

[mcp_servers.paretostudio.env]
PARETOSTUDIO_API_KEY = "psk_your_key_here"

Restart after every config change

Most MCP clients do not hot-reload. After editing the config, quit and relaunch Claude Desktop, reload Cursor / Claude Code, or start a new Codex session.

Verify install

After restarting your client, confirm ParetoStudio is wired up. Each tab below shows the success signal for one client.

Start a new Codex session in your project and ask “list my ParetoStudio prompts”. A library response, including an empty library, confirms the server is connected.

Then install a skill with client: "codex". It should appear at .agents/skills/<name>/SKILL.md.

First command

Ask your AI: “list my prompts”. If it returns your library (or an empty list for a new account), MCP is working.

Available tools

Twenty-one tools across read, install, write, and delete. Ask your AI client in natural language — it picks the right tool automatically.

Read

9 tools· Browse, fetch, and search every item in your library — personal and team.

`list_prompts`	List all prompts (personal ∪ team). Each row carries a [personal] or [team: org-id] scope marker.
`get_prompt`	Fetch a prompt and optionally resolve its {{variable_name}} placeholders with the variables argument.
`search_prompts`	Search prompts by title, description, or content.
`list_agents`	List all agents (personal ∪ team). Each row carries a [personal] or [team: org-id] scope marker.
`get_agent`	Fetch an agent configuration; pass variables to resolve Smart Variables for this use.
`search_agents`	Search agents by name or description.
`list_skills`	List all skills (personal ∪ team). Each row carries a [personal] or [team: org-id] scope marker.
`get_skill`	Fetch a skill configuration; pass variables to resolve Smart Variables for this use.
`search_skills`	Search skills by name or description.

Read tools surface personal and team content together

List and get tools automatically include items from every team org you are an active member of. Each row in the response is annotated with a scope marker: [personal] or [team: org-id]. Solo users see only personal rows — zero behavioral difference.

Install

3 tools· Drop agents and skills into a local project.

`install_agent`	Install under .claude/agents/. Parametrized agents require complete variable values before any file is written.
`install_skill`	Install SKILL.md for Claude or Codex. Parametrized skills require complete variable values.
`install_bundle`	Install an agent and multiple skills in one operation — full workspace ready.

Write

6 tools· Create and update library items from your IDE — within your plan's library limits.

`create_prompt`	Create a new prompt in your library.
`update_prompt`	Update an existing prompt and regenerate its formatted output.
`create_agent`	Create a new agent in your library.
`update_agent`	Update an existing agent.
`create_skill`	Create a new skill in your library.
`update_skill`	Update an existing skill.

Delete

3 tools· Remove items from your library. The client confirms before deletion.

`delete_prompt`	Delete a prompt from your library.
`delete_agent`	Delete an agent from your library.
`delete_skill`	Delete a skill from your library.

MCP works on every plan

Any plan with MCP access can browse, search, install, create, update, and delete. Creates count against your library limits (Free: 5 prompts / 2 agents / 2 skills; Pro and Team: unlimited). See rate limits for per-plan request caps.

Usage examples

Ask your AI client naturally. Here is what a few common requests look like and which tool they resolve to.

Browse

“List all my agents”

What the AI does behind the scenes

list_agents()

Find something specific

“Search my prompts for code review”

What the AI does behind the scenes

search_prompts({ query: "code review" })
// → returns matches sorted by relevance

Resolve a reusable prompt

“Get my launch brief for founders in a direct tone”

What the AI does behind the scenes

get_prompt({
  prompt_id: "pr_01H...",
  variables: { audience: "founders", tone: "direct" }
})

Install to a project

“Install my Backend Engineer agent into this project”

What the AI does behind the scenes

install_agent({
  agent_id: "ag_01H...",
  target_path: ".",
  variables: { stack: "TypeScript" }
})

Bundle agent + skills

“Install my QA agent with the testing and bug-report skills”

What the AI does behind the scenes

install_bundle({
  agent_id: "ag_01H...",
  skill_ids: ["sk_...", "sk_..."],
  path: "./"
})

Create from your IDE (Pro)

“Save this system prompt as a new agent called API Reviewer”

What the AI does behind the scenes

create_agent({
  name: "API Reviewer",
  system_prompt: "..."
})

Rate limits

Requests are capped per plan and reset every 24 hours. Hit the cap and the API returns 429 with a Retry-After header.

Plan	Daily requests	Tool access
Free	50	Read + Install + Write (capped at 2/2)
ProRecommended	500	All 21 tools (read, install, write, delete)

Troubleshooting

npx not found, hangs, or fails to fetch

The MCP config uses npx paretostudio-mcp@latest — that requires Node.js 18+ and a reachable npm registry.

Run node --version — upgrade if below 18.
Run npm config get registry — should print https://registry.npmjs.org/. On a corporate proxy, point it at your mirror or set HTTPS_PROXY.
First invocation downloads ~500 KB and can take 10–20s on slow links — let it finish before restarting the client.

Server loads but 0 tools appear

The MCP process started, but the client shows zero tools. Three usual suspects:

API key revoked or rotated — open Settings → Developer and confirm the key is still listed.
API key is from a Free account that has hit the 50-request daily cap. Upgrade in Settings → Billing for 500/day.
Missing psk_ prefix or a stray space when you pasted — re-paste the full key into env.PARETOSTUDIO_API_KEY and restart.

PARETOSTUDIO_API_KEY not set

The key must live in the env block of your MCP config, not in args. After saving the config, fully quit and reopen the client — most MCP clients do not hot-reload configuration.

Tools not appearing

Restart your AI client after editing the config. In Claude Desktop, quit and relaunch. In Cursor, reload the window (Cmd+Shift+P → Reload Window). In Claude Code, re-open the session.

API key not found or revoked

The key has been rotated or deleted. Generate a fresh one in Settings → Developer, copy the full psk_ prefix, and paste it back into your MCP config.

Rate limit exceeded (429)

You hit the daily request cap. The response includes a Retry-After header with the seconds until reset. Upgrade your plan for a higher cap.

Permission denied (403) on create / update / delete

Write tools are available on Free (capped at 2 agents / 2 skills) and unlimited on Pro. If you see 403 on a create/update/delete tool, you have hit the Free entity cap — upgrade in Settings → Billing.

list_prompts only shows personal items — team items missing

The union read requires active team membership. If you were invited but have not yet accepted, or if you were removed from the org, team rows are excluded. Accept the invite at /accept/[token] and then retry. See the Team guide for the full invite flow.

Back to Homepage

Developers

MCP Server

Prefer a guided setup?

Use the setup wizard at /setup/mcp. It picks your client, creates an API key, and verifies the connection in four steps. Continue reading if you want the manual walkthrough.

Prerequisites

ParetoStudio account

Any plan including Free. Free gets 50 requests/day; Pro unlocks 500/day and unlimited saves.

API key

Generate one in Settings → Developer. Starts with psk_.

Node.js 18+

npx paretostudio-mcp@latest fetches the latest bundle. No global install required. View on npm →

Setup

Paste the ParetoStudio MCP block into your client's config, fill in your PARETOSTUDIO_API_KEY, and restart the client. Pick the tab that matches where you work.

Config path: ~/.codex/config.toml

Add this block to your existing config without replacing unrelated settings. Start a new session after saving. Install native project skills with client: "codex".

~/.codex/config.tomltoml

[mcp_servers.paretostudio]
command = "npx"
args = ["-y", "paretostudio-mcp@latest"]

[mcp_servers.paretostudio.env]
PARETOSTUDIO_API_KEY = "psk_your_key_here"

Restart after every config change

Most MCP clients do not hot-reload. After editing the config, quit and relaunch Claude Desktop, reload Cursor / Claude Code, or start a new Codex session.

Verify install

After restarting your client, confirm ParetoStudio is wired up. Each tab below shows the success signal for one client.

Start a new Codex session in your project and ask “list my ParetoStudio prompts”. A library response, including an empty library, confirms the server is connected.

Then install a skill with client: "codex". It should appear at .agents/skills/<name>/SKILL.md.

First command

Ask your AI: “list my prompts”. If it returns your library (or an empty list for a new account), MCP is working.

Available tools

Twenty-one tools across read, install, write, and delete. Ask your AI client in natural language — it picks the right tool automatically.

Read

9 tools· Browse, fetch, and search every item in your library — personal and team.

`list_prompts`	List all prompts (personal ∪ team). Each row carries a [personal] or [team: org-id] scope marker.
`get_prompt`	Fetch a prompt and optionally resolve its {{variable_name}} placeholders with the variables argument.
`search_prompts`	Search prompts by title, description, or content.
`list_agents`	List all agents (personal ∪ team). Each row carries a [personal] or [team: org-id] scope marker.
`get_agent`	Fetch an agent configuration; pass variables to resolve Smart Variables for this use.
`search_agents`	Search agents by name or description.
`list_skills`	List all skills (personal ∪ team). Each row carries a [personal] or [team: org-id] scope marker.
`get_skill`	Fetch a skill configuration; pass variables to resolve Smart Variables for this use.
`search_skills`	Search skills by name or description.

Read tools surface personal and team content together

Install

3 tools· Drop agents and skills into a local project.

`install_agent`	Install under .claude/agents/. Parametrized agents require complete variable values before any file is written.
`install_skill`	Install SKILL.md for Claude or Codex. Parametrized skills require complete variable values.
`install_bundle`	Install an agent and multiple skills in one operation — full workspace ready.

Write

6 tools· Create and update library items from your IDE — within your plan's library limits.

`create_prompt`	Create a new prompt in your library.
`update_prompt`	Update an existing prompt and regenerate its formatted output.
`create_agent`	Create a new agent in your library.
`update_agent`	Update an existing agent.
`create_skill`	Create a new skill in your library.
`update_skill`	Update an existing skill.

Delete

3 tools· Remove items from your library. The client confirms before deletion.

`delete_prompt`	Delete a prompt from your library.
`delete_agent`	Delete an agent from your library.
`delete_skill`	Delete a skill from your library.

MCP works on every plan

Usage examples

Ask your AI client naturally. Here is what a few common requests look like and which tool they resolve to.

Browse

“List all my agents”

What the AI does behind the scenes

list_agents()

Find something specific

“Search my prompts for code review”

What the AI does behind the scenes

search_prompts({ query: "code review" })
// → returns matches sorted by relevance

Resolve a reusable prompt

“Get my launch brief for founders in a direct tone”

What the AI does behind the scenes

get_prompt({
  prompt_id: "pr_01H...",
  variables: { audience: "founders", tone: "direct" }
})

Install to a project

“Install my Backend Engineer agent into this project”

What the AI does behind the scenes

install_agent({
  agent_id: "ag_01H...",
  target_path: ".",
  variables: { stack: "TypeScript" }
})

Bundle agent + skills

“Install my QA agent with the testing and bug-report skills”

What the AI does behind the scenes

install_bundle({
  agent_id: "ag_01H...",
  skill_ids: ["sk_...", "sk_..."],
  path: "./"
})

Create from your IDE (Pro)

“Save this system prompt as a new agent called API Reviewer”

What the AI does behind the scenes

create_agent({
  name: "API Reviewer",
  system_prompt: "..."
})

Rate limits

Requests are capped per plan and reset every 24 hours. Hit the cap and the API returns 429 with a Retry-After header.

Plan	Daily requests	Tool access
Free	50	Read + Install + Write (capped at 2/2)
ProRecommended	500	All 21 tools (read, install, write, delete)

Troubleshooting

npx not found, hangs, or fails to fetch

The MCP config uses npx paretostudio-mcp@latest — that requires Node.js 18+ and a reachable npm registry.

Run node --version — upgrade if below 18.
Run npm config get registry — should print https://registry.npmjs.org/. On a corporate proxy, point it at your mirror or set HTTPS_PROXY.
First invocation downloads ~500 KB and can take 10–20s on slow links — let it finish before restarting the client.

Server loads but 0 tools appear

The MCP process started, but the client shows zero tools. Three usual suspects:

API key revoked or rotated — open Settings → Developer and confirm the key is still listed.
API key is from a Free account that has hit the 50-request daily cap. Upgrade in Settings → Billing for 500/day.
Missing psk_ prefix or a stray space when you pasted — re-paste the full key into env.PARETOSTUDIO_API_KEY and restart.

PARETOSTUDIO_API_KEY not set

The key must live in the env block of your MCP config, not in args. After saving the config, fully quit and reopen the client — most MCP clients do not hot-reload configuration.

Tools not appearing

Restart your AI client after editing the config. In Claude Desktop, quit and relaunch. In Cursor, reload the window (Cmd+Shift+P → Reload Window). In Claude Code, re-open the session.

API key not found or revoked

The key has been rotated or deleted. Generate a fresh one in Settings → Developer, copy the full psk_ prefix, and paste it back into your MCP config.

Rate limit exceeded (429)

You hit the daily request cap. The response includes a Retry-After header with the seconds until reset. Upgrade your plan for a higher cap.

Permission denied (403) on create / update / delete

list_prompts only shows personal items — team items missing

Back to Homepage