iBlueprint
Find, run, debug, and edit your iBlueprint Blueprints, use and save prompts, answer questions from your knowledge bases, plan project goals and tasks, and chat with your agents, from Claude, ChatGPT, or Codex. Ask for a Blueprint by name, and the assistant checks which inputs it needs, runs it, follows the run, and reports the result. It can also explain why a run failed, or design a change to a Blueprint and save it once you agree. It can fill in and use a prompt from your prompt library, save new prompts to it, and it can search or read the documents in your knowledge bases and answer with the sources it used. It can set goals in your projects, break them into tasks, keep those tasks up to date, and pass messages to your iBlueprint agents and relay their replies.
What you need
- An iBlueprint account. The plugin signs in to it in your browser and can see and do only what your account can in the app.
- AI connections and compute budget in iBlueprint for the Blueprints you run. Runs use them just as runs started in the app do.
Connect
When the assistant first uses iBlueprint, your browser opens the iBlueprint sign-in page and then asks you to allow the connection. In Claude Code you can start this yourself: run /mcp, select iblueprint, and choose Authenticate. To disconnect, go to Settings → OAuth → Connected apps in iBlueprint and select Disconnect.
What it accesses
The plugin contains no code that runs on your machine. It has two parts:
- Skills (
skills/iblueprint,skills/setup): instructions that tell the assistant to run a Blueprint only when you ask, and only once. They also tell it to ask for missing inputs rather than invent them, and to show you an edit before saving it. - The iBlueprint MCP server at
https://api.iblueprint.ai/api/mcp. Every tool call goes there, authenticated as you. The tools read your organizations, Blueprints, prompts, projects, goals, tasks, agents and their conversations, tools, run history, and knowledge bases, including the text of their documents. They change something only when the assistant calls a write tool:run_blueprint,apply_blueprint_operations,create_blueprint,create_prompt,update_prompt,create_project,create_goal,update_goal,create_task,update_task,add_task_note, orchat_with_agent. An agent you message can act through its own tools, such as sending email. Nothing can be deleted through the plugin. Saved prompts are private unless you share them with an organization. Before an edit is saved, the current Blueprint is stored as a version you can restore.
Nothing is sent anywhere other than iBlueprint. See the privacy policy, the terms of service, and support.
Install
Claude Code
claude plugin marketplace add iBlueprintAI/iblueprint-plugins
claude plugin install iblueprint@iblueprint
claude.ai and Cowork: add iBlueprint from the directory, or add the connector yourself under Settings → Connectors → Add custom connector with the URL https://api.iblueprint.ai/api/mcp.
Tools
| Area | Read | Write |
|---|---|---|
| Account | get_profile, list_organizations | |
| Blueprints | list_blueprints, get_blueprint_inputs, get_blueprint, get_node, list_node_types, get_node_type, validate_operations, and lookups of agents, tools, MCP servers, knowledge bases, and environment variables | create_blueprint, apply_blueprint_operations |
| Runs | get_execution, list_executions | run_blueprint |
| Prompts | list_prompts, get_prompt | create_prompt, update_prompt |
| Knowledge bases | list_knowledge_bases, list_knowledge_base_documents, get_knowledge_base_document, search_knowledge_base | |
| Projects | list_projects, list_goals, list_tasks, get_task | create_project, create_goal, update_goal, create_task, update_task, add_task_note |
| Agents | list_agents, get_agent_conversation, list_agent_conversations | chat_with_agent |
Development
| Path | Used by |
|---|---|
.claude-plugin/plugin.json, .mcp.json | Claude: Claude Code, claude.ai, and Cowork |
plugin.json (Agent Plugins manifest), mcp.json | ChatGPT and Codex |
skills/ | both |
evals/ | claude plugin eval; not loaded at run time |
tests/ | node --test, checks the portable package |
From the repository root:
claude plugin validate --strict iblueprint
node --test iblueprint/tests/package.test.mjs
claude plugin eval iblueprint --runs 1 --ablation none
The evals answer from mocked tools in evals/mocks/, so they need no iBlueprint account or network access.
License
MIT. See LICENSE.