Skip to main content
Treat Tiro not as just a meeting recorder but as the thing that feeds your AI agent, and the story changes completely. Here are prompts, grouped by situation, that let an agent read what you said out loud, organize it, and act on it. Use the copy button on the right of each code block and paste it as is. Replace bracketed placeholders like <topic> or <customer> with your own.
MCP and CLI are included on every plan at no extra cost. They work during your free trial too.

Connect Tiro first

Before any prompt works, connect Tiro to your agent once. Both routes are a single paste, after which the agent installs and signs in on its own.

If you have meetings often

Once meeting records pile up, they become an asset right away.

Use the Tiro CLI in Claude Code and Codex

Extract action items and run them
Apply meeting decisions to the code
Check what you dropped

Use Tiro MCP in Claude Desktop and ChatGPT

Build reporting material
Draft a weekly update
Trace how a decision moved
Turn a customer meeting into a proposal outline
Draft the follow-up email
MCP can look up a share link that already exists. Creating one is done in the app or through the API.
See who took on what

If you rarely have meetings

Tiro earns its place even without meetings. The point is to say things out loud instead of typing them.

Turn videos and talks into raw material

Play a video on your desktop app and let Tiro capture it. Tiro picks up the sound coming out of your computer, so you can do something else while it records. Say your own thoughts out loud as you go and the video and your commentary land in the same note.
Capturing your computer’s sound requires the desktop app. On Mac, grant the system audio recording permission once. See Recording video call and media audio for details.
Separate the video’s points from your own
Read several videos as one topic
Turn what you watched into writing

Code by talking

Turn Tiro on and let the ideas spill out. It feels odd the first time, then it clicks. Brainstorming flows far better than it does while typing, and you get many more ideas out.
This only needs your microphone, so you can try it on the web with nothing installed.
Build what you just said
Turn rambling into a spec
Talk your way through a stuck problem

Digest lectures, podcasts, and audiobooks

Build up research notes

Say each thought out loud as it comes while you read, and later an agent pulls the scattered notes into one thread. No repository needed, so this works even if you never touch code. Gather everything you have said on one topic
Keep only the questions you have not answered

Draft content

Something you said once can be reused in several shapes. Hand over a batch of notes and the agent finds the threads worth pulling. Find story ideas across scattered notes
Outline a series quoting only what you said

Run your own retrospective


What makes prompts work better

Name the note and the results improve. If the agent keeps grabbing the wrong note, run tiro notes list --limit 5 to see the list first and point it at a specific note. The full command reference is in the CLI quickstart.
If you want to block guessing, say so. Adding one line, “if it is not in the note, do not make it up, mark it as not in the note”, makes the output far more trustworthy.
Narrowing the time range sharpens the answer. Phrases like “the last 2 weeks”, “yesterday”, or “this month” shrink the pool of notes the agent digs through, so fewer irrelevant ones slip in.
Registering names, companies, and jargon before you record makes names come out far more accurate. See Note context and Wordbook.

Frequently asked questions

Do I need a paid plan to use these prompts? No. MCP and CLI are included on every plan at no extra cost, and they work during your free trial. Can I use these without a coding agent like Claude Code? Yes. Connect MCP to Claude Desktop or ChatGPT and most of the “If you have meetings often” prompts on this page work as is. Only the prompts that edit a repository need a coding agent. The agent says it cannot find my notes. Check that the connection is still live. Then narrow the scope by putting a time range or a note title directly in the prompt, which resolves most cases. If it still fails, see the search options in the CLI quickstart. Can the agent listen to the original audio? The agent only receives the transcribed text. To work with the audio file itself, use the API.

See also