Michael, let me help you. I have been in the "AI" industry for more than a
decade, and working with audio and signals as far back as building guitar
pedals at Caltech in the physics undergraduate lab. There was no requested
partnership between the FFMPEG group and the MCP Protocol in my emails, but
rather an application layer called an MCP Server, which is application
layer no different than building an FFMPEG wrapper in Python for
type-checking and user-proofing the unsigned, raw FFMPEG command line
interface, which I have also done. FFMPEG is painfully challenging to
script (or do you disagree, c'mon) and, yet, thousands of optimizations
(with gotchas) exist in its filter graphs (or, am I mistaken, and your
thirty years of development was not filled with thousands of hours of
message board discussions of complex audio video efficiencies and magic,
and for nothing more than a kiddie video concat device?). These are
precisely the logical structures LLMs (or, more appropriately, their CNNs)
can be trained to learn, and then implement through the MCP Server, for
accessibility reasons, for novice non-technical users to begin broaching
the deep technical expertise of the filter graph mechanisms, and also
through an expert software engineer designing an MCP Server to enable its
own structure checks, integrity checks, performance optimizations, and
more. Whether this compile interface business logic (as APIs, not unlike
that wrapper, juiced) is exposed as an MCP Server or a different framework
for interaction with LLMs specifically is moot, but this email exchange
suggests we are not even that far in understanding what "AI" and "hype
cycle AI" are, or why I felt your imperative to engage again, for your own
behalf. It is clear FFMPEG is not prepared, not seemingly interested in
expanding the FFMPEG user group or application of its most cherished
hard-fought domain expertise, and benefits from these emails more than I
currently benefit from sending them. It is recommended that we shelf this
issue for a month, while perhaps I build out a framework for my own uses,
and provide the FFMPEG group to discover what I am talking about before
they tell me what I am talking about. The earlier emails over the last two
years were so unproductive that I did not even both pushing my FFMPEG
wrapper from my private home server gitlab to the open source repository
where that will eventually go. Warm regards, Bernat

I suggest people interested email me directly, as flack is best suited for
outrunning air force jets, not messages from strangers.
[email protected]

PS Your tone, on top of your demonstration of my original inquiries,
considering *you do not know me*, is just weird.
It is truly disappointing to support open source these days.
https://github.com/davidbernat/FFMPEG


David Bernat, Ph. D.
Property of Starlight LLC.


On Sat, Jan 10, 2026 at 12:28 PM Michael Ivanov <[email protected]> wrote:

> David,  let me help you. Your excitement about MCP thingy is probably not
> shared by the people on this list, otherwise you would have already been
> replied. Here we discuss ffmpeg related issues. I can't see any
> relationship between this and the model context protocol. FFMPEG is for
> processing video and audio data, it is not an LLM video generator.  It has
> existed before the "AI" hype , and will be what it is after it is replaced
> by another one. I personally don't understand how MCP could advance this
> project. Please enlighten me.
>
> Best,
> Michael.
>
> On Sat, Jan 10, 2026, 19:19 David Bernat <[email protected]> wrote:
>
>> FFMPEG—
>>
>> 1. Is it possible none of the FFMPEG experts on this email list properly
>> know what MCPs do and what the state of the art of user-driven LLMs is
>> (especially in an AI expert partnered with a SME)?
>>
>> 2. Is it possible no work opportunities exist in this vast industry
>> within this team? It seems unwise.
>>
>> I replied twice. Figured I’d reply at the top and give y’all another
>> opportunity at a polite hello.
>>
>> Who with a decade of SME expert in FFMPEG does not reply to emails like
>> this, or delegate that to their intern for their intern to have a learning
>> opportunity?
>>
>>
>> Warm regards,
>> Bernat
>> #assemble 🇺🇸
>>
>> David Bernat, Ph. D.
>> Property of Starlight LLC.
>>
>>
>> On Sat, Jan 3, 2026 at 3:21 PM Michael Ivanov <[email protected]> wrote:
>>
>>> What should happen by 2033?
>>>
>>> On Sat, Jan 3, 2026, 18:07 David Bernat via Libav-user <
>>> [email protected]> wrote:
>>>
>>>> Hey esteem captains of FFMPEG,
>>>>
>>>> It is that awful topic every AI wanker (me) wings on about (AI coding).
>>>> I am a former chief scientist at AI companies, and have tried reaching out
>>>> through many channels over the last two years for mentors to better
>>>> understand exactly how FFMPEG wants to take itself into 2033 and beyond. I
>>>> am a non-power user audiophile with a background in physics and signal
>>>> processing.
>>>>
>>>> So, I just downloaded OpenCode this week to explore vibe-coding, and
>>>> ordered a V100 for $900 to set up an offline on-prem server. I spent a few
>>>> hundred hours in 2022 and 2023 building my own FFMPEG wrappers for Python
>>>> to produce various podcasts and videos for a YouTube channel. I am probably
>>>> your typical non-power user who wants to go that next level.
>>>>
>>>> An FFMPEG MCP is a unique challenge. Data processing is time consuming,
>>>> so any MCP would need to be equipped with extra protections of "explaining
>>>> AI" to walk through each and every filter suggested, and "sample AI" to run
>>>> these filters on small test clips to confirm each compiles (or video
>>>> inspection matches expectation) to run in a typical closed loop creation.
>>>>
>>>> That is the part that really excites me about joining any project. The
>>>> intricate systems of interacting filters is, of course, a defining asset of
>>>> FFMPEG and a critical headache for new users. And for power developers on
>>>> the FFMPEG team, this kind of critical un-black-box-ify is precisely what
>>>> in my experience is powerful to preserve. To learn this over the upcoming
>>>> year is to open the door to users (and me) learning to be power developers
>>>> nearer to you on the FFMPEG team. It also templates for a large swatch of
>>>> other types of code that need MCPs.
>>>>
>>>> I hope that you will take me under your wing with whatever working
>>>> group you know exists.
>>>> If this is better suited for a different email discussion list then my
>>>> apologies, and if any of you have received emails from me before asking for
>>>> contract opportunities you can reach out too.
>>>>
>>>> It would be *wonderful* to achieve sophisticated designs on par with
>>>> TikTok using fast, easy, and efficient LLM interactions with FFMPEG, for
>>>> the entire broader developer community. It would unbox FFMPEG for being the
>>>> drivers of these platforms themselves, and I am almost certain there are
>>>> cognate kids and developers to you all that would love to build that suite
>>>> over an upcoming year for their portfolios, job applications, own startups,
>>>> and more, starting here.
>>>>
>>>> Warm regards,
>>>> Bernat
>>>>
>>>> David Bernat, Ph. D.
>>>> Property of Starlight LLC.
>>>>
>>> _______________________________________________
>>>> Libav-user mailing list -- [email protected]
>>>> To unsubscribe send an email to [email protected]
>>>>
>>>
_______________________________________________
Libav-user mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to