Repository navigation
Releases
0.2.2
Takes musicfit 0.2.3, with guides that say what it does with a video that is not length long and what to ask the user for when it cannot read a track, a voice or a video, and easycaptionkit 0.1.5, with the captions guide's lines for its new libass stops.
When fit_music's video is not length long, to a frame, musicfit 0.2.3's reason gives only a length that length takes: 3600 for a video at most a frame past the hour, and 1 for one at most a frame under a second. The music guide says to give length those seconds, where it said to give the video's own length, which length refuses past the hour or under a second. Two new lines answer musicfit's two new refusals: for a video more than a frame past the hour, ask the user for a cut of an hour or less, with length set to its length, and for one more than a frame under a second, ask for a longer cut. fit_music's length, analyze_music's Next line and the music and checks guides ask for the video's length to a frame, where they asked for its exact length.
On the hosted server the line for a video past the hour asks for a cut a job there takes: 20 minutes at most with a video, 12 with a voice too, and less with more sound. Where musicfit's reason gives more seconds than that, the guide asks for a shorter cut as well. The service's own refusal, "fit_music makes a fit only as long as its video", for a video whose picture runs past the hour by what its file states, a Matroska or WebM picture counting as long as the whole file, moved from the list of parameters to that line, beside musicfit's, with the same answer. The guide says that a job's track, voice and video's own sound may run 15 minutes together at most, each as long as its file states, as the service counts them, without the reason of memory it gave, which musicfit 0.2.3 no longer bears out for a video's own sound: it reads that sound only up to two seconds past the end of the video's picture, which the fit matches to a frame.
When musicfit 0.2.3 cannot read a track, a voice or a video, the music guide's list of errors says what to ask the user for. After "Cannot read :" and why, as "it is cut short or damaged" or ffmpeg's own words, it is a whole copy of that file or one in a common format such as WAV, MP3 or MP4, a copy without the tag for "it starts with an ID3v2 tag", and one whose picture starts with its sound for "ffmpeg finds no size for its picture". For "ffmpeg reads no sound in " and "ffmpeg finds no sound stream in ", which the guide quotes where it quoted " has no audio stream", it is another file, and for "ffmpeg finds no picture in
On the user's own machine a file behind two or more ID3v2 tags can be taken for another kind, and a video can then read as another length than its own and be refused with that length to use: before giving length the seconds a refusal gives, the assistant asks the user whether the video is that long, and asks for a copy without the tags when it is not, or when a video's own sound, a track or a voice reads as none though the user says it has sound. There, when a call is refused for a line break in a path it gives, the assistant gives the path without it, and asks the user to rename the file or the folder only where its name holds the line break. The warning that the voice runs past the end is quoted so that it holds musicfit 0.2.3's "more than a second".
With fit_music held back, the troubleshooting guide gives analyze_music the same reasons for a track in musicfit 0.2.3's words, where it quoted musicfit 0.2.2's.
A folder named after a file, like musicfit-<track's name>, now takes _ for each line break, and each other C0 or C1 control character, in the file's name, so its name holds no line break. On the user's own machine a call whose path, a file's or out_dir, holds a line break or a C0 or C1 control character now starts nothing, as makersmcp's own lines print a path as it is: the refusal names the parameter and the character, and says to give the parameter without it, with the parameter's description for a text like cuts, and to rename the file or the folder, or copy the file, where a name holds it; for out_dir it says to give another or leave it out, and only to leave it out where the call cannot take one. The README, the guide to jobs and files and the troubleshooting guide name the paths makersmcp's own lines still print as they are, which no call gives: the folder makersmcp saves in, and in a refusal a file a film.json names or a file or a folder in a brand/ folder. A line break in one of those names starts a line of its own there, so the guides have the assistant ask the user to start makersmcp in a folder whose path holds none, or to rename that file or folder.
On the user's own machine, caption_video with libass true now stops before it draws where this ffmpeg's libass cannot use, or is not given, a file of the font kit given as fonts for the weight and style the captions set letters in, as easycaptionkit 0.1.5 does. The libass parameter and the captions guide say that drawing letters from a font kit's WOFF2 files with libass needs an ffmpeg whose FreeType reads WOFF2, which a static build of BtbN's FFmpeg-Builds does not, and the guide's line for a Chrome that does not start says libass can draw the captions without it. A new line of the guide answers "This ffmpeg's libass cannot use" and "libass cannot be given": the assistant asks the user for another file of the font, such as a TTF or OTF file, or, where the file is WOFF2 and this ffmpeg's FreeType reads no WOFF2, for an ffmpeg whose FreeType does, in FFMPEG_PATH, and where Chrome starts, calls caption_video again without libass. libass is a parameter of the user's own machine only, so jobs on the hosted server are not touched.
The lock takes reframekit 0.1.4, which a fresh npx makersmcp has had through its range since it came out.
No new tool, parameter or setting. mcp.makersmcp.com takes this release once api.makersmcp.com runs musicfit 0.2.3.
0.2.1
Fixes to the texts of the kits that joined in 0.2.0, and hosted texts that promise only what the service does.
A launch film with the user's track under it is reframed from the video.mp4 that fit_music made, with the film's film.timeline.json as cuts, so every place keeps the track. The tools and guides now say what the kits do: take_screenshots with pages 1 takes url and each page, a render from an edited crop.json is given the run's other arguments again, and the out_dir of reframe_video and plan_reframe must be empty or hold an earlier run of the same video.
FFPROBE_PATH is a setting for reframekit and easycaptionkit as well as launch-film. check_setup shows the ffprobe in use while any of them is on, and while the setting names no program, no tool of reframekit or easycaptionkit but list_caption_styles starts, unless an ffmpeg setting brings the ffprobe beside that ffmpeg.
On the hosted server the guides give the times and bounds measured there: a video too large for a job, or too long even as easy footage, is refused before its kit starts, while footage with grain can still be stopped at the job's time, a fit under a video is as long as the video's picture by what its file states, or as the whole file for Matroska and WebM, so a video past the hour takes no fit there, a length musicfit refuses is refused as the call is made, a length in frames under a video needs fps, and a job's links work, and its files go into a later job, until the time its result gives, which each result and each file parameter now says. A person's jobs run there one at a time, so the launch kit checks each job's files right after it, starts a step that needs an earlier job's file right after that job, and says that a step made again takes earlier jobs' files only within their time too. A service that runs makersmcp can read TEXTS_LEAVE_OUT, whose set, delete and clear throw, and textsLeaveOut for the tools it may hold back while the rest of their kit stays on, and MAX_REPEAT_RECORDS beside REPEAT_TIMING for how many calls the repeat ledger keeps, and a server it builds again in the same process for the same kits and texts lists each kit tool as it made it the first time. A kit tool call identical to one that ended in an error after its job started continues that job, as one cancelled or cut off does, instead of starting a second. mcp.makersmcp.com takes this release once api.makersmcp.com runs those checks.
0.2.0
makersmcp now carries all seven kits. reframekit (reframe_video, plan_reframe, check_place_limits) and easycaptionkit (caption_video, make_subtitles, list_caption_styles) join easybrandkit, easyfontkit, easyscreenshotkit, launch-film and musicfit, each at its newest release.
makersmcp_playbook gives an assistant a guide for each kind of job, and every tool and result says when to use it and what comes next. A call with an argument a tool does not take is refused before anything starts, an identical call continues the job it already started, and what a kit prints is shown as data.
The hosted MCP server at https://mcp.makersmcp.com/v1 runs the kits too, with sign-in through Google, GitHub or an email code; reframekit and easycaptionkit join it soon. See https://makersmcp.com/setup/.
The repository moved to github.com/MakersMCP/makersmcp, and makersmcp is on the MCP Registry as io.github.MakersMCP/makersmcp.
0.1.1
0.1.0
The first release on npm. makersmcp is one MCP server for five free kits that run on your own machine: easybrandkit, easyfontkit, easyscreenshotkit, launch-film and musicfit. They all come with it, so npx -y makersmcp@0.1.0 is the whole install, and the README has the setup for each app. This first version was published from a local session, so it carries no provenance; the next ones go out through GitHub Actions with trusted publishing.