IterPrompt MCP

• Search the public library of skills, prompts, and templates.
• Read your saved and private packages — the same library across every agent.
• Use IterPrompt prompts through MCP’s native prompt interface.
• Ask for package recommendations for the task at hand.
• Save useful packages to your library.
• Create new private skills from things they learned.
• Fork and improve existing skills, with attribution.
• Submit improvements for human review — nothing goes public automatically.
• Report whether a skill worked, improving future rankings.

Open Dashboard → MCP for step-by-step instructions for your client and your personal server URL. In short: add the IterPrompt server URL as a remote MCP server in your client; the client sends you to IterPrompt to sign in and approve the permissions you want to grant. Clients that can’t use OAuth can use a personal access token instead (Connections → “Create access token” — an advanced credential shown exactly once; treat it like a password).

Connections start with read/save access. Anything that writes to your library requires explicit approval on the consent screen:

• library:read

  Search the public library and read packages you have access to

• library:save

  Save public packages to your personal library

• packages:write elevated

  Create private packages, fork packages, and edit your packages

• submissions:write elevated

  Submit your packages for public review

• outcomes:write elevated

  Report task outcomes after using a package

• profile:read

  Read your basic profile (name and email)

Dashboard → MCP → Connections lists every connected client with its scopes, last-used time, and recent actions. Revoking a connection immediately invalidates all of its credentials — the very next request fails. You can also rotate personal access tokens there.

With packages:write approved, an agent can save a skill it developed during a task. The package is created private, as a draft, marked as MCP-created, and content-scanned for secrets and unsafe instructions. It appears under My Packages where you can edit it — every save creates a new immutable version.

Private packages reach the public library only through review: draft → validation → submission → pending review → approved or rejected. You (or an agent with submissions:write) submit a specific version; a human reviewer sees the content, diffs, attribution, and validation findings before deciding. Track status under Submissions.

• Review third-party skills before use. A skill is a set of instructions your agent will follow — it can tell the agent to take actions. Automated scanning catches obvious problems but does not guarantee safety.
• Tokens are stored only as hashes and shown once at creation. Never paste them into prompts, logs, or shared configs.
• The activity log stores sanitized summaries only — never your repository contents, task context, or full prompts.
• Outcome statistics are self-reported by agents and are not verified evaluations.

• 401 / authentication required: the token expired or the connection was revoked — reauthorize from your client, or create a fresh access token.
• missing_scope errors: reconnect and approve the elevated permission the tool needs.
• Rate limit reached: wait a minute; limits are per-connection and per-operation type.
• Client can’t connect: run “Test MCP connection” on the setup page — it checks the endpoint, authentication, and tool listing and tells you which step failed.