SocialKit·the v2 engine, live

The AI engine
for social.

It writes the post, renders the image on brand, judges its own work, and revises until it clears the bar. Then it publishes on time and reads the results back. One API, every feed.

Made by the engine

real renders · zero templates

Every image here came out of the API from a brief and a brand kit. There is no template gallery behind it: the engine deals a layout archetype, picks a typography voice, renders deterministically, and a judge grades the result before you ever see it.

Post the engine rendered for Kiln House Ceramics
Kiln House Ceramicssplit-blockArchivo Black
Post the engine rendered for Driftwood Roasters
Driftwood Roasterssplit-blockOswald
Post the engine rendered for Vela Studio
Vela Studiophoto-fullPlayfair Display
Post the engine rendered for Driftwood Roasters
Driftwood Roastersphoto-fullOswald
Post the engine rendered for Vela Studio
Vela Studioposter-gridPlayfair Display
Post the engine rendered for Kiln House Ceramics
Kiln House Ceramicstext-panelArchivo Black
Slide from a carousel deck the engine rendered
Driftwood Roastersslide 1 · hook
Slide from a carousel deck the engine rendered
Driftwood Roastersslide 3 · step 1
Slide from a carousel deck the engine rendered
Driftwood Roastersslide 7 · proof

It deals whole decks too: these three slides are from one eight-slide carousel, one frame family, roles anchored per slide. Fictional brands. Real, unretouched engine output. Each artifact is its own world: images, carousels, video.

One engine, the whole loop

brief to feed and back again

SocialKit grades drafts against the real algorithm, writes in your voice, renders the visual on brand, judges the design, revises what falls short, publishes exactly once, and reads engagement back to grade smarter. One bearer key over REST or MCP. Built to live inside an agent loop. See how it works.

01Score

Grade any draft against the live platform rubric. A number, a verdict, and the exact fixes.

02Generate

Turn a brief into ready-to-post drafts in your brand voice, each auto-graded best-first.

03Render

Design the image too: a layout archetype, a typography voice, your palette and fonts. Deterministic pixels, not a prompt lottery.

04Judge

A design judge grades every render, TEMPLATE to GREAT, and puts its reasons on the record.

05Revise

Weak posts get fixed, not shipped. Feedback becomes a surgical spec change: the photo stays, the headline moves.

06Publish

Queue and pace posts with webhooks, retries, and idempotency keys. Exactly once, on time.

07Measure

Pull real engagement back, per post and account-wide, straight from the platform.

08Calibrate

Predicted score meets actual reach, and the gap feeds the next grade. The loop closes.

Pixels are a spec

deterministic render · revisable

JSON in, pixels out.

One call takes a brief and a brand kit and returns a finished post: caption, hashtags, rendered masters, and the judge's score. Behind every image is a serializable spec. The same spec renders the same pixels, an edit is a diff, and a revision can fix the headline without touching the photo.

~/socialkit · curl
bash
# A finished post, image included
curl https://api.socialkit.sh/v1/generate-visual-v2 \
  -H "Authorization: Bearer $SOCIALKIT_KEY" \
  -d '{
    "businessContext": {
      "businessName": "Driftwood Roasters",
      "brandKit": { "palette": { ... }, "fonts": { ... } }
    },
    "seed": { "platform": "instagram", "idea": "the Tuesday roast" }
  }'

# → { caption, hashtags,
#     masters: [{ aspectRatio: "4:5", imageUrl, ... }],
#     score: { verdict, score, singleFix, ... },
#     spec: { ...the revisable recipe } }

Every feed, one contract

2 live · 3 coming

Each platform is graded against its own 2026 rubric, so the same call returns LinkedIn-smart or X-smart output. More feeds are landing on the same contract.

LinkedInliveXliveInstagramsoonFacebooksoonTikToksoon

Tools · PostScore

Free, no signup
MR
Maya ReevesFounder, Quietloop
I rejected a $400k contract last week. Best decision I've made all year. Here's why slowing down is the most underpriced growth tactic in B2B right now: the deals you say no to teach you more about your ICP than the ones you close.
LinkedIn324 charsFounder voice
87/100
Verdict

Strong hook, clean POV. Tighten the middle and this lands in the algorithm's top decile.

Hook23 / 25
Algorithm fit18 / 20
Specificity14 / 15
Structure11 / 15
Voice14 / 15
Engagement7 / 10
+Hook lands in 7 words. Specific number ($400k) earns dwell.
!Middle paragraph runs long. Break the 32-word sentence into two.
!No question or CTA. Engagement signal is weak.
Score your own post

The API

60+ operations · self-serve

The whole engine, one key.

Score, rewrite, plan, generate, and the visual engine, plus the brands and voices that condition them, posts with a real lifecycle, channels, webhooks, and metrics. Built to live inside a loop, not a chat window.

Read the docs
~/socialkit · curl
bash
# Score a draft
curl https://api.socialkit.sh/v1/score \
  -H "Authorization: Bearer $SOCIALKIT_KEY" \
  -d '{
    "post": "We cut onboarding from 11 fields to 3 and activation jumped 40%."
  }'

# → { overall: 87, breakdown: { ... } }

For agents

MCP · 36 tools, annotated

Let your agent manage your social.

SocialKit speaks MCP and returns structured signals, designed for a draft, grade, rewrite loop. Every tool carries honest annotations: read-only ones an agent can run freely, destructive ones flagged for confirmation, idempotency keys where it writes. Drop it into Claude, Cursor, or your own runtime.

SocialKit for agents
~/.claude/mcp.json
json
{
  "socialkit": {
    "url": "https://mcp.socialkit.sh",
    "auth": { "bearer": "$SOCIALKIT_KEY" }
  }
}

// 36 tools: score_post, rewrite_post, validate_post,
//   generate_post, plan_week, build_voice,
//   generate_visual, revise_visual,
//   generate_carousel, revise_carousel,
//   brands + voices CRUD, create_post, update_post,
//   schedule_post, bulk_schedule, publish_post,
//   cancel_post, delete_post, get_calendar,
//   score_saved_post, refresh_metrics, get_analytics,
//   get_calibration, and the rest of the loop.

// read-only tools are annotated safe to auto-run;
// destructive ones ask first.
The proof

SocialKit is the engine. Taka is the proof it works.

Taka is our own social media manager for small businesses, and every post it ships comes out of this engine: the words, the pixels, the schedule, the metrics. The API is the same door we use ourselves. Two brands, because the people who care how the feed works want a different door from the people who want their social done for them.

Start now

free to start · no card · no waitlist

Mint a key. Make your first call.

No sales call, no approval queue. Sign in to mint a key from the dashboard, or POST your email from the terminal and get a live key back. Either way you score your first post in under a minute, and agents can sign themselves up the same way.

Free starts at 100 credits a month, no card. A credit is one successful model call, charged only when it works. Upgrade when you outgrow it.

~/socialkit · curl
bash
# 1. Get a key
curl -X POST https://api.socialkit.sh/v1/account \
  -H "content-type: application/json" \
  -d '{"email": "you@company.com"}'
# → { "apiKey": "sk_live_..." }

# 2. Score your first post
curl https://api.socialkit.sh/v1/score \
  -H "Authorization: Bearer sk_live_..." \
  -d '{"post": "We cut onboarding from 11 fields to 3."}'