Skip to content

Releases: getsquish/squish

v0.2.3 — discoverable: MCP Registry listing + the video-navigation skill

Choose a tag to compare

@v1b3x0r v1b3x0r released this 14 Jul 05:05
abc35eb

The idea: distribution shifted from "apply" to "be discoverable" — this release is about agents finding Squish on their own: an official MCP Registry listing, an installable navigation skill, and an agent-facing decision page in the docs.

In 0.2.3 (over 0.2.2):

  • Listed in the official MCP Registrypackage.json now carries mcpName: io.github.getsquish/squish (the registry's npm ownership marker), server.json describes both ways in (the hosted endpoint + this npm package), and a tag-triggered OIDC workflow publishes the listing. One listing, two transports.
  • video-navigation skillnpx skills add getsquish/squish teaches an agent the whole practice in one install: when to reach for a contact sheet, the zoom loop, and the citation discipline (never cite finer than the cell spacing; confirm before citing). Live on skills.sh.
  • stdio MCP truthfulnessserverInfo now reports the real package version (it had sat at a hardcoded 0.2.0 through two releases), and squish_video carries behavior annotations: writes sheets to your disk, nothing leaves the machine (closed world), reruns overwrite the same outputs (idempotent).
  • Docs for agents — new For AI agents page: a pick-your-surface decision table, 8 behavior rules for reading and citing a sheet, quota/refusal handling, and the machine-readable consumption paths. Plus a proper Support page.
  • The hosted endpoint (same engine, not in this package) became a ChatGPT app candidate — Apps SDK metadata (outputSchema + structuredContent, attachment fileParams), a subject-keyed free lane, and a commerce-free quota voice for ChatGPT traffic. Submitted for review 2026-07-12.

Engine unchanged: same frozen squish-cli-v0 / squish-mcp-v0 contracts; the zoom loop since 0.2.0.

Install: npx -y @getsquish/squish <video> · MCP: npx -y @getsquish/squish mcp in any stdio client · Node ≥ 20, ffmpeg on PATH.

Made at getsquish.app 🦑

v0.2.2 — the lens README + the remote mouth grows auth

Choose a tag to compare

@v1b3x0r v1b3x0r released this 10 Jul 13:37
9442f47

The idea, sharpened: AI sees through lenses, not answers — Squish adjusts the lens; the model interprets. The window (--start/--end) is the lens made wide or narrow, density is the lens made coarse or fine, and the navigation loop is the lens moved until the answer is observable.

In 0.2.2 (over 0.2.1):

  • README rebuilt around the lens — new cover ("Random access to video"), the official-app path (Claude Desktop / claude.ai → add https://api.getsquish.app/mcp as a connector, zero install), and fresh field proof: the same task run across 5 clients and 3 mouths in one day; Claude Desktop completed the multi-round navigation loop on its own, down to a sub-second lock.
  • squish_video description tuned from field evidence — the one-call economy line ("one call replaces a whole download → ffmpeg → extract → montage pipeline"), density-vs-window guidance, and the timecodes[][] mirror note. Agents pick the tool and drive the loop better with no prompting changes on your side.
  • The hosted endpoint grew auth (same engine, not in this package): api.getsquish.app/mcp now takes an optional Authorization: Bearer API key — same credits as the hosted API, deduct-before-extract with auto-refund — alongside a small anonymous free lane; running out answers with a structured error carrying billing_url that an agent can relay in conversation. The local CLI and MCP server in this package stay ungated forever.

Engine unchanged: same frozen squish-cli-v0 / squish-mcp-v0 contracts; the zoom loop shipped in 0.2.0.

Install: npx -y @getsquish/squish <video> · MCP: npx -y @getsquish/squish mcp in any stdio client · Node ≥ 20, ffmpeg on PATH.

Made at getsquish.app 🦑

v0.2.1 — first open-source release

Choose a tag to compare

@v1b3x0r v1b3x0r released this 08 Jul 08:31
005e4c6

The engine behind @getsquish/squish is now open source — Apache-2.0, mirrored from the founding monorepo, history starts here.

The idea in one line: give AI random access to video — instead of watching a clip linearly, an agent navigates it: overview → zoom (--start/--end) → cite absolute timestamps.

In 0.2.1 (over 0.2.0, the last npm-only release):

  • README: the meta-demo (▶ watch the 76-second explainer — or read the same video as one contact sheet), the primitive framing, and the navigation loop
  • package.json repository now points at this repo — npm's "Repository" link finally works
  • SECURITY.md (private reports via GitHub advisories) + npm / CI / license badges

Engine unchanged: same frozen squish-cli-v0 / squish-mcp-v0 contracts; the zoom loop shipped in 0.2.0.

Install: npx -y @getsquish/squish <video> · MCP: npx -y @getsquish/squish mcp in any stdio client · Node ≥ 20, ffmpeg on PATH.

Made at getsquish.app 🦑