Deep research that questions the question
Give Tin a project question. It tests the question and the assumptions underneath it against current, source-backed evidence, and saves a report you can build on.
codex mcp add tin --url https://app.tin.computer/mcpWhat the workflow does
A Tin workflow is a repeatable job: declared inputs, a bounded run and a defined output. The steps below are what this one does, every time.
- 01Takes the question, depth, audience and what you already assume.
- 02Identifies the upstream assumptions the question rests on.
- 03Gathers current evidence with sources.
- 04Saves reports/RESEARCH_DEEP_DIVE.md to the project.
Have Tin research whether usage-based pricing fits our product. List the assumptions you want it to test.
What every workflow here comes with
Battle tested on real products
These are the workflows we run for real companies. Claw Messenger went from about $45 to $2,105 in monthly revenue in 124 days on them. Each one has declared inputs, bounded execution and tests, and anything public waits for your review.
See the case study →Open source: read it, fork it, self-host it
Apache 2.0. Every workflow is a folder in the repository: the steps it runs, the data it reads and what it is allowed to do. Change it, or run the whole system on your own machine.
Open the repo →Runs on the agent that already knows your product
Research is only as good as the framing. Your agent knows your constraints and frames the question around them.
All workflows →Workflows that run well with this one
Questions
How is this different from a chat assistant's research mode?
The report lands in your project, where later workflows and your agent can use it, and the method is open source.
Can I control how deep it goes?
Yes, with the depth input.
Is Tin really open source?
Yes. Tin is licensed under Apache 2.0 and the source is at github.com/tin-computer/tin. You can read every workflow, change it, and self-host the whole system.
Which coding agents can run Tin workflows?
Any agent that speaks MCP, including Claude Code, Codex and Cursor. There is also an API, and the same projects run in the browser at app.tin.computer.
Why run marketing from a coding agent?
Because it already knows your product. It has the repository open, so it fills workflow inputs accurately, reviews output against what the product really does, and can act on the results in the same session.