Ozmen
@nozmen
Aqua Voice: I barely type long prompts now, just talk. Saved ~5 hrs typing in 2 weeks. Much more accurate than macOS dictation.
Claude, ChatGPT, Cursor, Codex, VS Code, Windsurf, Zed: if it speaks remote MCP with OAuth, it can call Aqua. One URL, no per-client setup.
Ten tools, each one narrow verb. Read tools return what you already have in Aqua. Write tools change your account, and your client asks before the first one runs. There is no admin scope and no access to anyone else’s transcripts.
“Summarize my last three brainstorm dictations from this week and pull out anything that sounds like a hiring commitment.”
“Read the transcript of yesterday's customer interview, draft a two-paragraph follow-up in my voice, and file it as a Linear ticket.”
“When I dictate to you and Aqua mishears Baseten as Base 10, fix it and teach Aqua the correction automatically.”
“Add every product code name in this design doc to my Aqua dictionary, and turn ML into machine learning only when casual messaging is off.”
The dictionary, writing rules, and language you set in Aqua are the same ones your agent reads and edits over MCP. Corrections it files train your future dictations.
Connect your agent
Connect your agent
Session feedback

When Aqua mishears a name, your agent files feedback on that session. Aqua learns the fix for every dictation after it.
Custom dictionary

Names, brands, product codes. Your agent adds them atomically against a revision number, so two clients never overwrite each other.
Writing style

Casual on Slack, structured in docs. Your custom instructions live in the same payload as the dictionary, so an agent can read them and revise them.
49 Languages

Flying to Berlin? Ask your agent to set German. It changes one setting and leaves the rest exactly as they were.
One URL. Any client that speaks remote MCP. No password form.
One URL for every client that speaks remote MCP. It is the card just below.
Claude Desktop: Settings, then Connectors. Cursor: Settings, then MCP. ChatGPT, Codex, VS Code, Windsurf, and Zed use their own remote-MCP flow.
Google or Apple, same as the app. Read access is approved on connect. Write is a separate approval the first time an agent needs it.
https://api.aquavoice.com/mcp
Then ask your agent to call aqua_get_me. It should name your account and your privacy mode back. Full tool reference in the API docs.
Every dictation app puts words on the clipboard. Only one of them is also a server your agent can talk to. Wispr Flow, Spokenly, and Apple Dictation list no MCP server. Aqua ships a remote one, with OAuth, today.
Read the full head-to-heads: Aqua Voice vs Wispr Flow, vs Spokenly, and vs Apple’s built-in dictation.
Used by people at
Tulsi Prasad ⚡️
@thebuildguy
Unpaid shout out — @aquavoice is probably one of my most used tool that has saved me quite literally a day of time!
jaxn
@jackson_felty
@aquavoice is a prime example of how much better a product can be when the team focuses on making one simple use case incredible
juan
@juanbuis
this is why @aquavoice is better than all other AI transcription apps it actually writes like you talk
mass
@Memetic_Theory
If you're not using @aquavoice you're moving 15% slower than you could
Ferruccio Balestreri
@0xferruccio
For coding prompting with voice nothing beats @aquavoice All the other transcription apps don't work well with code like tailwind class names
Pablo Antonio
@PabloAntonio
Besides Manus, Cursor, and others besides Gemini, there are other AI tools I like, but the one tool I absolutely can't cancel is Aqua Voice
Nathan Ruff
@TheNathanRuff
Been using @aquavoice for a few months. Unbelievably better than the other companies that do this
turboblitz
@turboblitzzz
Started using @aquavoice a few weeks ago and now I barely type anything. My housemates want to kick me out because i keep talking to my computer. Life is good.
chirag
@chiggyAI
just tried out @aquavoice and by far the most accurate voice to text tool i've used!!
utku
@utkusays
just subscribed to aqua voice and wow you guys are cooking, best dictation app i've tried so far
Individuals
Get started for free
1,000 free words
Avalon transcription model
Unlimited, tuned dictation
Unlimited words
Custom Instructions
Expanded Custom Dictionary
AI power-user features
Everything in Pro
See words live with Realtime Mode
“Send it” voice command
Early access to new features
Organizations
For teams of 2-9
Everything in Pro
Centralized billing
Team-wide settings
Enforce Privacy Mode
For teams of 10+
SSO/SAML + SCIM
Advanced Reporting
Zero Data Retention
Team-wide Dictionaries
Volume Discounts
Aqua MCP is included on every plan. 1,000 free words is about 7 minutes of talking at a normal pace. No credit card required.
No. That is Aqua typing into Claude's input for you. This is Claude reaching back into your Aqua account to read past dictations, correct itself, and change your settings. Both are useful. Together they close the loop.
The tools you approve, and nothing else. Read scope covers list, get, get-transcript, customizations, and settings. Write scope, requested separately, covers feedback and changes to customizations and settings. There is no admin scope and no access to other people's transcripts.
No. Every session an agent can see belongs to the connected user.
Any client that supports remote MCP with OAuth. Claude Desktop, Claude Code, ChatGPT (with Apps developer mode), Cursor, Codex, VS Code with the MCP extension, Windsurf, and Zed.
The transport returns 403 with an insufficient_scope challenge naming the exact scope missing. Most clients prompt to re-authorize instead of failing opaquely.
Feedback filed through aqua_send_session_feedback is used to correct your own future dictations. See the Trust Center for the full data policy.
Limited access today. Your AI provider must support remote MCP and may require workspace admin approval. Email support@aquavoice.com if you want in before it's open to everyone.