Exact agent prompt
Use Proforma's native MCP tools to help me with this task. Read https://proformalabs.ai/skill.md for product guidance.
Help me get started in Proforma. Continue any task or research already supplied in this conversation. If I am unsure, get to know my work casually, one simple question at a time, and suggest a few useful first steps for me to choose across sales, marketing, product or operations. Help me analyze data, build an App, automate work or test an idea, then build on what I learn. For a chosen experiment, define a useful comparison, success measure, guardrail and review date; keep the resulting learning and next decision connected. Use another method when the question calls for it. A company research deck is an option, not a requirement. Reuse my existing account, Session, evidence and saved work. Use only live advertised capabilities and needed access, keep credentials private, distinguish assumptions, observations and conclusions, and return saved Proforma links. Do not assume a specialist workflow or perform unrequested external effects.
1. Reuse my existing Proforma MCP connection. If native Proforma tools are unavailable, show https://app.proformalabs.ai/mcp and https://app.proformalabs.ai/docs/platform-mcp; ask me to configure one Streamable HTTP server in my host and complete browser OAuth for the intended Workspace. Use either an installed Proforma plugin or a direct MCP registration, not both. Do not edit host configuration or collect credentials in this conversation. Resume in a chat where the tools are available; a setup link is not a connected host.
2. Use proforma_whoami when Workspace, scopes or Model policy are unknown. Use proforma_search_tools with a focused task query and a bounded result limit when the operation is unknown. Before supplying summarized nested inputs, retrieve the exact tool's complete schema with includeSchemas: true. Reuse known complete contracts and inspect existing artifacts before creating anything. Ask only for the access needed. After a scope or Model-policy denial, explain the missing permission; never switch to CLI, HTTP, another connection or delegated Chat to bypass it.
3. Start from the task and existing work supplied here. Keep research and execution in this conversation unless I request delegated Proforma Chat and grant chat:run. For a template, inspect proforma_list_packages and proforma_get_package, then use proforma_install_package to install or reuse its complete reviewed package, preserving Models, Actions, Sources and owned state. Retain the package version, App link, destination database and installed Action IDs. Do not clone only the App or silently replace customized artifacts.
4. Return saved App links and exact Connection, Action, execution, revision and snapshot receipts supplied by the tools. Distinguish accepted, queued, pending, continuing and completed work. Follow existing execution receipts instead of duplicating work. Disclose synthetic records and missing evidence. Copying this prompt does not install a package, authorize a provider or run an Action; claim those states only after successful execution and readback.
How the connection works
Authorize the Workspace. Keep the task.
Add the MCP URL in your host and complete browser authorization for one Workspace. The host retains its credentials. Approve only the scopes and Model access needed; you can revoke access in Settings. Copying the prompt or creating an account does not install or run the workflow.
- 01you
Create your account Sign up or sign in through the secure link.
- 02agent
Picks up the context Continues your task or helps find a useful first step.
- 03you
Choose a first step Start with your work, data or an idea worth exploring.
- 04agent
Helps you start Begins the project you choose, with only the access it needs.
What the agent can do
Files and Apps
Create and update durable work with version history and links people can open.
Data and operations
Find approved data, run queries, use automations, and configure tracking when you allow it.
Ways to connect
Use native MCP to read, build and run approved work, or choose HTTP for an HTTP-capable agent.