# Stackcut

> Stackcut finds the software subscriptions a founder or small team pays for and shows cheaper ways to get the same job done: switch vendor, a cheaper plan, open source, pay per use, or replace the part they use with AI (with a build spec).

## For agents

- [MCP server](https://stackcut.io/mcp): Streamable HTTP, stateless, no auth. Tools: get_my_plan, start_audit, choose_stack, plan_stack_from_spec, list_stack_jobs, audit_stack, get_usage_questions, get_detection_rules, search_recipes, get_recipe, find_alternatives, get_build_packet, search_workflow_recipes, get_workflow_recipe, leaderboard, list_categories, get_exit_plan, search_exits, prepare_negotiation, report_action_outcome, report_exit_outcome, report_outcome. Resources: stackcut://exits and stackcut://exits/{id} (the cancel library). Also at https://stackcut.io/api/mcp. Setup: https://stackcut.io/agents
- [MCP server, directory endpoint](https://stackcut.io/mcp/public): the same tools except get_my_plan; routine calls retain no arguments or answers; explicit action reports retain anonymous structured outcomes for review.
- [Server card](https://stackcut.io/.well-known/mcp.json): name, version, both endpoints, how many tools each has, docs and contact (support@stackcut.io).
- From a web page: the catalog tools answer any site. report_outcome, report_exit_outcome, report_action_outcome and get_my_plan answer MCP clients (they send no Origin header) and pages on stackcut.io only; a request with another site's Origin gets HTTP 403.
- The user's own plan: get_my_plan with the read-only agent code (sca_…) they copy from stackcut.io/app → Send to my agent. Every line carries its proof (receipts, sourced prices) and steps; savings that can't be proven are null. Ask the user before any cancel or purchase.
- Choosing what to build with ("best database for this", "cheapest way to build X"): choose_stack with the jobs and expected volumes (MAU, emails/month, events/month, tokens, images, GB stored, video minutes, transactions and more; several per job). Add scales for cost curves and break-evens. A tier is a fit only when its published limits cover the volume; blocked_by says why a tier doesn't fit (daily caps, commercial use, session limits). Hand a product brief to plan_stack_from_spec and it works out the jobs and volumes, with editable assumptions.
- Baselining a team's stack before a decision (cut spend across tools, consolidate apps, replace one tool): start_audit with a rough manifest, or just vendors and team_size. It returns the missing facts that would change the answer, highest stakes first, each with why and where to find it; collect only those, list what you can't find in unknowns, and call again until status is 'enough', then call next.tool (audit_stack) with next.arguments. Manifest schema: https://stackcut.io/schemas/stack-manifest-1.json. Send counts, feature names, prices and dates, never names, emails or secrets.
- Auditing what someone pays for: audit_stack or find_alternatives with price, seats, features and usage per service (or evidence strings like '40,000 emails/mo', 'db 1.5 GB'). A path is eligible only when every check passes; otherwise the answer lists the checks and the questions to ask. get_usage_questions says which numbers decide each vendor and where the user finds them. Right-size paths find the cheapest plan of the same vendor that fits. With the invoice, pass billing (scope, account_id, the avoidable monthly charge with avoidability_confirmed, commitment months left, source): only what leaving removes from the bill counts, after the commitment ends.
- A job they don't pay for yet: search_recipes with intent "new" lists each recipe's options with what they cost (monthly, setup, first year, and the usual paid tools at list price), never savings against a tool they don't have. With requirements (must_have, exclude, budget_monthly_usd), options of an excluded kind are left out and each must-have is labeled supported, not_supported or not_checked. A search that names two or more tools ("replace Notion, Asana and Trello with one tool") comes back with a consolidation hint: start_audit, then audit_stack.
- Migrating a whole workflow (hosting, email, auth, queues, storage, AI routing and more): search_workflow_recipes, then get_workflow_recipe for a reviewed runbook with the evidence to collect, decision paths, acceptance tests, rollout, rollback and starter templates. Runbooks are instructions, not results.
- Cancelling, pausing or paying less for any subscription (streaming, music, news, dating, cloud storage, AI tools, apps, software): get_exit_plan with the service (plan words are fine: "Spotify Premium Family", "Canva Pro"), who bills them (billed_by) and the renewal date. It returns every route with steps and the words to expect on screen, when access ends, notice, fees and refunds, what to export first, cheaper ways to keep it, and agent rules: the person signs in, sees any offer and confirms the final click. With renewal_date, last_safe_day counts back the documented notice (days, business days, months or hours) plus a day. With monthly_price and renewal_date (or months_left), a fee stated as a share of what's left comes back as early_termination_estimate, an estimate: the vendor's bill decides. search_exits browses the library by name, domain, category or a word from the summary; report_exit_outcome counts what happened (no personal data). The [stackcut-leave skill](https://stackcut.io/downloads/skills/stackcut-leave.zip) walks an agent through a cancellation with these tools.
- billed_by accepts exactly these values: `direct_web`, `direct_app`, `apple_app_store`, `google_play`, `amazon`, `roku`, `paypal`, `carrier_or_isp`, `reseller_or_partner`, `sales_contract` and, for get_exit_plan only, `unknown` (every route and the question to ask). These aliases are read as their channel, in any case and with spaces or hyphens: `apple`, `app_store`, `ios`, `itunes` (apple_app_store); `google`, `android`, `play` (google_play); `carrier`, `isp`, `phone`, `cable`, `tv_provider` (carrier_or_isp); `partner`, `bundle`, `reseller`, `marketplace` (reseller_or_partner); `web`, `website`, `direct`, `online`, `company` (direct_web); `app`, `in_app` (direct_app); `contract`, `sales`, `enterprise`, `invoice` (sales_contract).
- Stored data: Stackcut stores the arguments of tool calls (redacted, never IP addresses) to improve recommendations. Don't send personal data or secrets.
- [Recipe catalog (JSON)](https://stackcut.io/data/recipes.json): every recipe, replacement path, modeled cost and source; where available, how to leave the current vendor (exits), step-by-step playbooks per path, and Stackcut's own end-to-end test reports (lab).

## Pages

- [Cancel library](https://stackcut.io/cancel): how to cancel any subscription, one page per service (e.g. [Netflix](https://stackcut.io/cancel/netflix)), written from each company's own help pages with a quote and check date for every fact. Open data, CC BY 4.0: [index.json](https://stackcut.io/exits/index.json), [all.json](https://stackcut.io/exits/all.json), [schema](https://stackcut.io/exits/schema.json).

- [Home: find a tool by the job](https://stackcut.io/): type the job ("a team wiki for 8 people", "a kanban board") and see every way to get it done (a cheaper vendor, open source, an agent skill, build it yourself), priced for the team size against what you pay now, with the math behind each number. The jobs index is [data/home-jobs.json](https://stackcut.io/data/home-jobs.json); each option's `m` holds the inputs of its year-one number and [lib/math.js](https://stackcut.io/lib/math.js) re-costs them.
- [Free stack audit](https://stackcut.io/audit): find every subscription in your receipts (Chrome extension or card CSV) and see what to cancel, pay per use or replace.
- [Blog](https://stackcut.io/blog): what to use instead of Semrush, Ahrefs, Typeform, Zendesk/Intercom, Loom, Slack, Notion, Airtable, Salesforce/HubSpot, Mailchimp, Zapier, Calendly, Docusign, Vercel, Supabase, Resend/SendGrid, ChatGPT/Claude and Cursor/Copilot, with sourced prices, how to leave, and Stackcut's own test receipts.
- [Head-to-head pages](https://stackcut.io/blog#head-to-head): one tool against the one people move to (e.g. [Calendly vs Cal.com](https://stackcut.io/blog/calendly-vs-cal-com), [Mailchimp vs Kit](https://stackcut.io/blog/mailchimp-vs-kit), [Semrush vs OpenSEO](https://stackcut.io/blog/semrush-vs-openseo)): prices from both vendors, what comparison pages claim labeled by who wrote them, what the importer carries over, public switching stories, and how to leave. Only research an independent checker confirmed against its source.
- [Use cases](https://stackcut.io/use-cases): the cheapest stack for common jobs (transactional email, database and auth for an MVP, SEO data for agents, a support inbox, a newsletter).
- [Replacement recipes](https://stackcut.io/recipes): searchable by vendor, feature or category.
- [Savings leaderboard](https://stackcut.io/leaderboard): paths ranked by modeled first-year net savings.
- [How it works](https://stackcut.io/how-it-works): the free stack audit (Chrome extension or card CSV).

## Notes

- Savings are modeled from public prices and planning assumptions (setup and upkeep valued at $50/hour), not measured results. Confirm must-have features before recommending a switch.
- Every modeled number can be shown step by step: year one = 12 × (what you pay now − the new cash cost − upkeep hours × $50) − (setup hours × $50 + setup cash). The MCP returns it as how_calculated on each path; recipe paths carry the inputs in `math`.

## Audit a codebase

- [Stackcut skill](https://stackcut.io/downloads/stackcut-skill.zip): SKILL.md + scripts/detect.mjs. Scans a repo locally for paid services (packages, env var names, config files, API calls) and sends only vendor and feature names to audit_stack.
- Stackcut's tested agent skills (Apache-2.0; each was run on 10 to 14 test cases, then re-run by a fresh agent and graded by a separate one; the write-up is on its recipe page):
  [meeting notes from a transcript](https://stackcut.io/downloads/skills/stackcut-meeting-notes.zip),
  [email drafts in your voice](https://stackcut.io/downloads/skills/stackcut-email-drafts.zip),
  [template images from data](https://stackcut.io/downloads/skills/stackcut-template-images.zip),
  [weekly KPI digest](https://stackcut.io/downloads/skills/stackcut-kpi-digest.zip),
  [sales pipeline in CSV files](https://stackcut.io/downloads/skills/stackcut-crm-files.zip),
  [propose meeting times by email](https://stackcut.io/downloads/skills/stackcut-booking-assistant.zip),
  [support reply drafts](https://stackcut.io/downloads/skills/stackcut-support.zip). They draft and prepare; a person approves anything that leaves the machine.
- [Detection rules (JSON)](https://stackcut.io/data/detect.json): about 100 vendors; also the MCP tool get_detection_rules.

## Negotiation and feedback

- `prepare_negotiation` prepares documented existing-customer, cancellation-flow and renewal routes with source dates, account eligibility questions, and draft request/terms/counteroffer/decline/confirmation scripts. It sends nothing and changes no account. Confirm the billing channel, market, full price, duration, later renewal price, fees and any new commitment. Unknown amounts remain null; cost comparisons are modeled.
- Structured pilot paths: Placeit, Norton, IONOS and TotalAV. A documented route does not guarantee an offer for an account.
- `report_action_outcome` accepts an explicit user-reported result tied to the packet revision and route ID. Reports stay unverified; blocked or changed steps create private review tasks. Reuse an opaque deduplication key when retrying. Keep receipts, email messages and personal data private. The legacy `report_exit_outcome` only counts outcomes.
- [Coverage and evidence counts](https://stackcut.io/exits/coverage.json): 462 English guides, 1,347 billing routes, 50 localized guides, 434 services with pay-less options, eight structured negotiation paths. Freshness has an as-of date; documentation checks and observed successes are distinct.

## Data

- [Price index (JSON)](https://stackcut.io/data/products.json): plans, published caps, overage and fees per vendor, each with its source and checked date.
- [Open-source alternatives (JSON)](https://stackcut.io/data/oss_catalog.json): forkable projects per vendor with license, stars and self-host requirements as observed.
- [Agent skills (JSON)](https://stackcut.io/data/skills_catalog.json): 51 skills, plugins and MCP servers that replace part of a paid workflow, with the license read from each repo, what they cover and the paid APIs they still need. On the site each one is a way to do a job (the agent-skill column), priced at $0 on top of the agent you already pay for, or marked "priced by usage" when it needs a paid API.