Loopcard competition — 18 September 2026

Product under test. Loopcard: a shareable coding-agent bill card (PNG + public /c/{id} + OG image). Preview: https://cryptobook.space. Not a gateway, not traces, not MCP security, not a Mac spend app.

Method. First-party product, pricing, GitHub, Product Hunt, and HN pages opened on this date. Secondary aggregators used only as pointers. Vendor “used by N developers” claims are labelled as vendor claims.

One-line verdict. Accounting for agent spend is solved and free. The default is npx ccusage@latest (18.6k GitHub stars) plus Anthropic/Cursor CSV exports. SuperPenguin (YC W26) already sells the paid team SKU at $0 / $30 / $200. Langfuse ($29) and Helicone (maintenance mode) own traces. ShieldMCP ($49 PDF, 5 PH upvotes) and MCP Observatory ($29 + $15k pilot) already proved that a pretty public card without a weekly job does not convert. Loopcard wins only if the object is a screenshot people post, not a ledger people keep. If the homepage looks like ccusage, Langfuse, SuperPenguin, or ShieldMCP, we are dead.

Category map

Four lanes. We sit in one cell. Every other cell is occupied.

                    LOCAL LEDGER                         TEAM ATTRIBUTION
              ccusage 18.6k ★  MIT $0               SuperPenguin YC W26
              CodeBurn 11.1k ★ MIT $0               $0 / $30 / $200
              cctally 8 ★ Apache $0                 Mac app + SDK, not a proxy
              claude-code-cost 7 ★ MIT $0
              (tables, TUI, localhost dashboard)

  VENDOR INVOICE (source of truth)                 PUBLIC SHARE OBJECT
  Anthropic Console CSV  (API)                     Loopcard  ← us
  Claude.ai Analytics CSV (Team/Ent)               Year in Code     HN 3 pts
  Cursor dashboard CSV   (Teams)                   Cursor Wrapped   weekend toy
  Claude Code /usage                               ccusage --compact (screenshot tables)
  (accurate, ugly, free, no OG)

  LLM TRACES / GATEWAY (do not enter)              MCP BLAST CARD (already killed)
  Langfuse Core $29  34.7k ★                       ShieldMCP $49 PDF, PH 5 upvotes
  Helicone $79  6.2k ★  maintenance                MCP Observatory $29 + $15k
  LiteLLM OSS 59k ★                                CodeFlowOps free 30s grade
                                                   (SSL Labs economics)

Named products (opened 18 Sep 2026)

1. Accounting CLIs and local dashboards

ccusage — the default, not a hole

Local CLI. Reads JSONL already on disk. No upload. Daily / weekly / monthly / session / 5-hour blocks / statusline. Unified reports across Claude Code, Codex, OpenCode, Amp, Droid, Codebuff, Hermes, pi, Goose, OpenClaw, Kilo, Kimi, Qwen, Copilot CLI, Gemini CLI, Antigravity, Grok Build CLI, ZCode. JSON export. --compact is explicitly “for screenshots and sharing.” Costs are estimates from token counts + LiteLLM pricing (cache-aware). MIT. npx ccusage@latest.

• GitHub: 18.6k stars, 840 forks — github.com/ccusage/ccusage
• Docs: ccusage.com · JSON: ccusage.com/guide/json-output
• npm: ccusage v20.0.22 published 17 Sep 2026 — npmjs.com/package/ccusage
• Product Hunt (8 Jul 2025, “How to use ccusage”): 109 upvotes, #17 of the day, 3 comments — hunted.space recap
• HN: no breakout Show HN. Used as a citation in spend threads (“ccusage says I burn $600 of API on a $100 Max”). Example: news.ycombinator.com/item?id=47701637
• Price: $0. Sponsored by Lineman.io (team Claude Code spend product).

Vs Loopcard: they already own the terminal screenshot. --compact is their PNG. We win only if a stranger clicks a URL. We lose every power user who already has npx muscle memory.

CodeBurn — the CLI we must not pretend does not exist

Not on the original named list. Omitting it would be malpractice. Free local tracker, MIT, 40 tools, TUI + web dashboard from localhost + macOS menu bar + desktop app + GNOME panel. Optimize tab for cache bloat / retry tax. Vendor claim: “150k+ developers.” Treat that as unverified marketing. GitHub is the traction we can count.

• GitHub: 11.1k stars, 839 forks — github.com/getagentseal/codeburn
• Site: codeburn.app
• Product Hunt: 106 upvotes, #11 of the day, 129 followers, 0 reviews, launched ~Aug 2026 — producthunt.com/products/codeburn
• Price: $0. MIT. No account.

Vs Loopcard: they have the local dashboard SuperPenguin-shaped feature set at $0. They do not host a public OG card. If they add “Download share card,” our wedge is a weekend.

cctally — ccusage-compatible, quota-native, almost no stars

Local Python CLI + live dashboard on 127.0.0.1:8789. Signature view: $/1% of weekly quota. Forecast, budget, desktop alerts, 5-hour blocks, Codex parity, conversation viewer. Drop-in cctally claude <cmd> for ccusage. Honest about accuracy: Claude dollars are transcript-derived lower bounds, not /usage billing totals. Apache-2.0. Opt-out telemetry beat.

• GitHub: 8 stars, 2 forks (v1.109.0, 15 Sep 2026, 1,814 commits — a lot of product, almost no audience) — github.com/omrikais/cctally
• npm: ~10.8k weekly downloads on a 23 Aug 2026 npm search snapshot — npmjs.com/search?q=keywords:quota-tracking
• Price: $0.

Vs Loopcard: they export shareable Markdown / HTML / SVG with project names anonymized. That is a local share object. Still not a public URL with OG. Stars say the market does not care about a better local dashboard once ccusage exists.

claude-code-cost — local PWA dashboard, 7 stars

npx claude-code-cost --open. Localhost dashboard. Drill-down total → project → session → message. Same pricing logic as ccusage (costUSD when present, dedupe by messageId+requestId). 17 themes, PWA-installable. MIT.

• GitHub: 7 stars, 1 fork — github.com/NikiforovAll/claude-code-cost
• npm: v0.23.0, ~1.3k weekly downloads (Snyk snapshot) — npmjs.com/package/claude-code-cost
• Demo: nikiforovall.blog/claude-code-cost
• Price: $0.

Vs Loopcard: pretty local charts. No public share URL. Themes are not a company.

Same name, different objects (do not confuse):

• github.com/keakon/claude-code-cost — Python ccc, 6 stars.
• roninforge.org — Claude Code plugin /claude-code-cost:report. Zero-install inside the harness.
• github.com/jbarbier/claude-code-cost-estimate — /cost-estimate skill for human rebuild cost, 109 stars. Different job.

2. Mac / team spend apps

SuperPenguin (Carrot Labs, YC Winter 2026)

AI spend intelligence. Not a proxy. SDK wraps the provider client; Mac app meters Cursor / Claude Code / Codex locally and syncs aggregates; GitHub connection prices merged PRs. Prompts off by default. 14 providers on the homepage (OpenAI, Anthropic, Gemini, Deepgram, ElevenLabs, Vercel, Bedrock, Azure, Modal, LiteLLM, Together, Fireworks, OpenRouter, Cursor).

Pricing, opened on superpenguin.ai 18 Sep 2026:

• YC: ycombinator.com/companies/carrot-labs — W26, team size 2, San Francisco. (YC launch post is a different Carrot Labs pitch — custom models. The live product at superpenguin.ai is spend attribution. Do not mix the two in marketing copy.)
• Docs: superpenguin.ai/docs — “SDK is not a proxy.”
• No public GitHub star count for the product (closed). No Product Hunt page found on this pass. OSS clone advertised against the $200 tier: github.com/officialasishkumar/costobs (X post 10 Jun 2026).

Vs Loopcard: they own the paid job (“what did this PR cost, across providers”). We are not a $30 FinOps tool. If we add “team dashboard” we become a worse SuperPenguin on a VPS.

3. Token cutters (adjacent, not the card)

Tokenade

Local hooks that compact noisy tool output / MCP results before the model reads them. Claims −38.9% mean cost on their own Token-Harness Optimizer Leaderboard (vendor benchmark, 7 long Claude Code tasks × 10 runs). Requires a free account even on Free. Pro is a savings cap, not a seat.

Pricing, opened on tokenade.net/pricing and tokenade.net/en/pricing:

• GitHub: 6 stars, 0 forks — github.com/pi-infected/tokenade-npm
• npm: @tokenade/cli v1.1.17, 15 Sep 2026 — npmjs.com/package/@tokenade/cli
• Positions vs ccusage: “ccusage measures, we reduce” — tokenade.net/en/comparisons/tokenade-vs-ccusage

Vs Loopcard: different job. If we claim we “cut your bill,” we are lying and we collide with them. Print the number. Do not sit on the agent’s I/O.

4. Vendor dashboards and CSVs (the invoice)

Anthropic / Claude

• API Console Usage + Cost pages: filter by workspace / model / key / month; CSV export of displayed data. Roles: Developer, Billing, Admin. Explicit limitation: cannot break down by individual user. support.claude.com — Cost and Usage Reporting
• Claude.ai Team/Enterprise Analytics: spend, tokens, top users, by product (Chat, Claude Code, Cowork, Office Agents). Download menu → Summary / Usage over time / By product / By model / Top users CSVs, or a ZIP of all. Spend report: per-user, per-model, up to 90 days, one-day lag. support.claude.com — View usage analytics · claude.com/docs/government/org-admin/analytics
• Claude Code /usage: the live quota the CLIs reverse-engineer. Transcript-derived tools (cctally, ccusage) admit they undershoot this.

Cursor

• Teams/Enterprise Usage Analytics at cursor.com/dashboard/analytics. CSV per chart, or header download of all charts. Filter: ≤10 users, ≤90 continuous days. Admin API: Enterprise only. cursor.com/docs/account/teams/analytics
• Individual Spending tab is the plan-allowance authority (included pool vs on-demand). Forum users already confuse “included $20 of API-equivalent” with a cash invoice — forum.cursor.com/t/cursor-billing-daily-usage-dashboard/150110
• Third-party CSV visualizers exist and stay local: github.com/jnst/cursor-usage (npx @jnst/cursor-usage, drag-drop, “export PNGs”); cursortokens.vercel.app; Cursor Wrapped (russ.dev/posts/cursor-wrapped) — “a spreadsheet isn't something you'd share,” then a weekend share-card. That sentence is our entire category, already attempted.

Vs Loopcard: the CSV is the product for anyone who will open a spreadsheet. We are a renderer. Cursor or Anthropic shipping “Share as image” is an existential 90-day risk. Mitigation: be the public URL + OG, not the parser.

5. Traces — why we are NOT them

Langfuse

Open-source LLM/agent tracing, prompts, evals, cost per trace. MIT core. Cloud:

Overage $8/100k units, down to $6/100k at 50M+. langfuse.com/pricing

• GitHub: 34.7k stars, 3.8k forks, release 17 Sep 2026 — github.com/langfuse/langfuse
• Logos on pricing: Canva, Twilio, Ramp, Khan Academy.

Why we are not Langfuse. They instrument before the fire, store spans, score generations, run experiments. Loopcard is a coroner PNG from a file the user already exported. No SDK. No traces. No prompts. No units. If we add a waterfall, Core at $29 is the ceiling we lose to, and the OSS Docker Compose is $0.

Helicone (Mintlify, maintenance mode)

Proxy-first LLM gateway + observability. Acquired 3 Mar 2026. First-party: “services will remain live for the foreseeable future in maintenance mode.” helicone.ai/blog/joining-mintlify. Historical: 14.2T tokens, 16k orgs, PH #1 (vendor claim in that post). Pricing still live:

helicone.ai/pricing

• GitHub: 6.2k stars, 670 forks — github.com/Helicone/helicone

Why we are not Helicone. Maintenance mode is a migration pool, not a vacancy. Cloning a base_url proxy is a BRIEF kill and a CFAA/key-custody kill. Their mental model is “see every request.” Ours is “one number, one picture, after the fact.”

6. MCP blast cards — the kill we already accepted

Listed because the operator will be tempted to “just add a grade.” Demand already said no.

ShieldMCP

Paste claude_desktop_config.json / Cursor / Windsurf → OWASP MCP Top 10 score in ~60s. Free score. $49 one-time PDF with copy-paste fixes.

• Product Hunt: 5 upvotes, 6 followers, 0 reviews, launched ~Jun 2026 — producthunt.com/products/shieldmcp
• Site: shieldmcp.net

That is the demand ceiling for “pretty report of what the agent can do.” Five people clicked upvote. Zero reviews. A $49 PDF is not $19/mo.

MCP Observatory (KryptosAI)

npx -y @kryptosai/mcp-observatory@latest. Local scan free. Hosted snapshot free (1 target, 7-day). Individual Pro $29/mo (10 targets, 90-day, hosted CI). Real SKU is $15,000 Release Gate Pilot (1–3 servers, 10 business days). Safety Index: 176 servers. Homepage still uses “Used by developers at Accenture / Cisco / Oracle” — logo-adjacent marketing we will not copy unless true.

• Site: mcp-observatory.com
• GitHub: 146 stars, 14 forks (releases page 18 Sep 2026); contributor file snapshots also show 174 — treat ~146–174. github.com/KryptosAI/mcp-observatory

Vs Loopcard: same object shape (public card, score, URL), wrong job (MCP security, not the bill). Their $29 has no indie conversion evidence we opened. Their money is a sales pilot. If Loopcard’s /c/ becomes a blast-radius grade, we are ShieldMCP with pink.

Feature table

Prices (self-serve, this date)

Psychological ceiling if we ever charge: under Langfuse Core $29, and under SuperPenguin Growth $30. $19 is the only number that is not a joke next to npx. There is still no evidence anyone pays $19 for a public bill card. ShieldMCP’s $49 PDF and Observatory’s $29 are the closest paid objects and they are ghosts.

Traction signals (countable, with URLs)

Read the table without romance:

• Stars cluster on local CLIs and on Langfuse. 18.6k + 11.1k people starred a terminal/TUI that never posts to X.
• Share-card Show HNs die at 1–3 points. Year in Code is Loopcard with a Spotify skin, fed by ccusage JSON, processed in-browser. HN shrugged.
• PH upvotes on cost trackers do not become reviews. CodeBurn #11, 0 reviews. ShieldMCP 5, 0 reviews.
• Paid WTP in this neighborhood is SuperPenguin $30 (teams) and Langfuse $29 (hosted ClickHouse). Not a PNG.

Where we uniquely win

Only three things incumbents will not ship this quarter because it fights their positioning:

1. Beauty as the product. ccusage is proud of Unicode tables. Anthropic CSV is a finance export. SuperPenguin is a workspace. None of them want a pink mugshot of a $6,000 overnight on the timeline. That would make their users look reckless, which is bad for a FinOps brand and irrelevant to a CLI.
2. A share URL a stranger can open with no account. /c/{id} plus og.png. ccusage --compact is a terminal screenshot you still have to crop. cctally SVG is local. Cursor Wrapped used html2canvas on a personal site. Year in Code made you upload JSON to yearincode.xyz and got 3 HN points — but they did not sit on a domain with a labeled public-story demo already filled in. The demo card (re-creation, labeled) is the distribution, not the parser.
3. OG image in the link unfurl. X/LinkedIn/Slack show the number without a login. Vendor dashboards require the session cookie. SuperPenguin requires the workspace. Langfuse requires the project. This is the only growth unit that matches the GTM (screenshot-first).

That is the whole wedge. It is a media object, not a system of record.

Honesty check: ccusage could add --png. CodeBurn already has a menubar and a PH gallery of charts. SuperPenguin could render a “this PR cost $18.40” card tomorrow. Anthropic could put “Share” on Analytics. The win is time-to-first-post and not being the accountant. The moment we chase accuracy, we lose to people who already have 18.6k stars and an invoice API.

Where we lose (do not argue)

Accuracy. cctally prints it on the README: Claude transcript totals are lower bounds, not /usage. ccusage is API-equivalent, not the Max-plan invoice. Anthropic CSV is the bill. Cursor Spending tab is the pool. Loopcard prints whatever is in the dropped file. If the file is an estimate, the card is an estimate. Marketing that implies we “match the invoice” is a lie next to Console export. Label it.

Local-first trust. ccusage/CodeBurn/cctally never need a server. Loopcard’s parse is in-browser (good). Publish sends aggregates to our VPS (necessary for OG, fatal for the paranoid). CLI users who will not POST a number will download a PNG and leave. That is the correct outcome. Do not add OAuth to “help” them.

CLI users. The person who already runs npx ccusage@latest every Monday does not need us. They have statusline hooks, 5-hour blocks, --json into jq, and a compact table for Twitter. Our buyer is the person who screenshots Anthropic’s dashboard after a scare, or who wants to post the $6k story. Overlap with ccusage’s 18.6k stargazers is a leak, not an ICP.

Team FinOps. SuperPenguin $30. Cursor Teams analytics. Claude Analytics CSV. We have no seats, no Slack alert, no cost-per-PR. Do not add them.

Live quota. Claude /usage, cctally forecast, CodeBurn menubar, CC Usage Bar. We are after-the-fact.

MCP / traces / gateway. Occupied. Killed. BRIEF.

Conversion. Closest paid analogs (ShieldMCP $49, Observatory $29) did not convert. Year in Code did not convert. A viral PNG is a Virtus Labs awareness play until /pricing shows money. Treat $19 Pro as upside, not the plan.

12-month competitive response if we get attention

Assume Loopcard cards actually hit X/HN. What incumbents do. Plan for it, do not be surprised.

What we should not do when copied: add accuracy features, MCP, traces, Electron, Stripe-before-virality, or “the only usage tracker.” The response that survives is: faster demo, sharper story, better OG, three parsers, no secrets.

What kills us even if we “win” the screenshot war: one-shot virality with no return path. The honest floor is Virtus Labs awareness. The dishonest plan is 150 × $19/mo from people who already have ccusage.

Implications for Loopcard (competition only)

1. Homepage job is “make the bill a card,” not “track spend.” Track is ccusage/CodeBurn/SuperPenguin.
2. Never compare on stars, parsers, or invoice fidelity. We lose.
3. Cite ccusage as the ledger. Docs already say this. Keep saying it. They are distribution, not an enemy.
4. Do not become SuperPenguin, Langfuse, Helicone, ShieldMCP, or Observatory. Those homepages are occupied and two of them already taught us the card does not swipe.
5. Expect --png from the CLI lane within a quarter of any traction. Our residual is the public URL + labeled $6k demo + OG. That is a media property. Budget it like one.

One sentence that still holds on 18 Sep 2026:

Loopcard is the shareable PNG and OG URL of a coding-agent bill, parsed in the browser from an export you already have — not ccusage, not SuperPenguin, not Langfuse, and not another MCP score.

If that sentence is not true of the live site, this file’s verdict is we already cloned a dead lane.

URLs opened (18 Sep 2026)

CLIs / local:

• https://github.com/ccusage/ccusage
• https://ccusage.com/
• https://ccusage.com/guide/
• https://www.npmjs.com/package/ccusage
• https://github.com/omrikais/cctally
• https://github.com/NikiforovAll/claude-code-cost
• https://www.npmjs.com/package/claude-code-cost
• https://nikiforovall.blog/claude-code-cost/
• https://github.com/getagentseal/codeburn
• https://codeburn.app/
• https://www.producthunt.com/products/codeburn

Mac / team / optimizer:

• https://superpenguin.ai/
• https://superpenguin.ai/docs
• https://www.ycombinator.com/companies/carrot-labs
• https://tokenade.net/pricing
• https://tokenade.net/en
• https://tokenade.net/en/pricing
• https://github.com/pi-infected/tokenade-npm
• https://www.npmjs.com/package/@tokenade/cli

Vendor CSVs:

• https://support.claude.com/en/articles/9534590-cost-and-usage-reporting-in-the-claude-console
• https://support.claude.com/en/articles/12883420-view-usage-analytics-for-team-and-enterprise-plans
• https://claude.com/docs/government/org-admin/analytics
• https://cursor.com/docs/account/teams/analytics
• https://github.com/jnst/cursor-usage

Traces / frozen gateway:

• https://langfuse.com/pricing
• https://github.com/langfuse/langfuse
• https://www.helicone.ai/pricing
• https://www.helicone.ai/blog/joining-mintlify
• https://github.com/Helicone/helicone

MCP cards (anti-pattern):

• https://www.producthunt.com/products/shieldmcp
• https://mcp-observatory.com/
• https://github.com/KryptosAI/mcp-observatory

HN / PH:

• https://news.ycombinator.com/item?id=46401120 (Year in Code, 3 pts)
• https://news.ycombinator.com/item?id=47284990 (CC Usage Bar, 1 pt)
• https://news.ycombinator.com/item?id=48005014 (WakaTime spend, 2 pts)
• https://news.ycombinator.com/item?id=47701637 (ccusage $600-on-$100)
• https://www.hunted.space/product/how-to-use-ccusage/launches/how-to-use-ccusage

Internal (this repo, not traction):

• /root/cryptobook-next/docs/03-competitive-landscape.md
• /root/cryptobook-next/docs/11-loopcard-product.md
• /root/cryptobook-next/research/competition.md
• /root/cryptobook-next/research/demand-who-pays.md
• /root/cryptobook-next/challenges/competition.md
• /root/cryptobook-next/challenges/slip-demand.md