Frequently asked questions
Updated 2026-06-06.
Product basics
Is SpecStep a code generator?
No. SpecStep generates a documentation package — the spec your build runs from: requirements, architecture decisions, acceptance criteria, a test plan, and an implementation sequence. It doesn't write your application code. It writes the plan your AI coder (or your team) builds from.
Why do I need a spec if I already use Claude Code, Cursor, or Codex?
AI coders are excellent at writing code and poor at deciding what to build. Without a clear spec they drift — inventing requirements, contradicting earlier decisions, and rewriting things you didn't ask them to touch. SpecStep gives the coder a stable, reviewed plan to follow, so it builds the thing you actually meant.
What does the generated package include?
A markdown bundle covering vision and scope, requirements, architecture decisions, acceptance criteria, a test plan, an implementation sequence, and supporting reference material — plus the review notes the specialist agents produced along the way. You can download it as a zip or pull it over the REST API or MCP.
Can an AI coding agent use the package directly?
Yes — that's the point. The package is written to be read by an AI coder. Pull it over MCP from inside Claude Code, Cursor, or Codex and the agent works straight from the plan, or hand it the downloaded files. It's also perfectly readable by humans.
Is the output legal, security, or compliance advice?
No. SpecStep's specialist reviews — including security, legal, privacy, and compliance notes — are informational quality aids produced by AI, not professional services, audits, certifications, or legal advice. Review the output and consult a qualified professional where it matters. See the Terms for the full disclaimer.
Pricing, credits & plans
What does SpecStep cost?
Free to start, then credit-based paid plans. SpecStep uses credits as the unit of generation; you spend credits by how deep a review you choose. The current plans, prices, and credit allotments live on the Pricing page so they stay in sync with the in-app picker.
What is a credit, and how many does each review depth use?
A credit is what one generation spends, scaled to its review depth: Fast = 1 credit, Normal = 2 credits, Extensive = 4 credits. Deeper reviews cost more because they run more specialist passes. Your plan includes a monthly credit allotment; the Free tier includes one Fast credit to try a real generation.
Do I need a credit card to try SpecStep?
No. The Free tier doesn't ask for payment information. You can run through an interview, spend your free credit on a real generation, and see what comes out before deciding whether a paid plan fits.
Can I buy one package without subscribing?
That's the plan — one-off generation packages (priced by review depth) let you buy a single spec without a subscription. One-off purchases and paid subscriptions become available as we enable payments; the Pricing page shows what's purchasable right now.
How do I cancel?
From Settings → Billing. You keep access through the end of the current billing period and aren't charged again. We don't pro-rate refunds for partial periods or for credits already spent.
Do unused credits roll over?
No. Your monthly credit allotment resets at the start of each billing period; unused monthly credits don't roll over. If you regularly run out, the next plan up — or a one-off package for an occasional extra spec — is usually the right call.
What happens if I run out of credits?
New generations pause until your credits reset, you upgrade, or (where available) you top up. Everything you've already made — packages, interviews, integrations — stays accessible; only new generations are blocked.
What happens if a generation fails or is cancelled?
A generation that fails before producing a package doesn't burn the credit. If a run is interrupted (for example, by a deploy), SpecStep resumes it rather than charging you twice. If something looks wrong with a charge, file a support ticket and we'll sort it out.
Generations & review profiles
Which review profile should I pick?
- Fast (1 credit) for prototypes, proofs of concept, and one-off scripts.
- Normal (2 credits) for production code your team will maintain. This is the default.
- Extensive (4 credits) for systems where the cost of getting it wrong is high — payments, security-critical paths, regulated workloads.
The interview's profile picker shows a per-intake credit cost and duration estimate that adapts as Otto learns more about your project — grounded in your actual usage history rather than fixed approximations.
How long does a generation take?
Roughly: Fast finishes in a few minutes; Normal takes around ten to fifteen; Extensive can run up to half an hour. The interview shows a live estimate before you commit.
Can I edit a generated package?
Yes — the package is yours. Download the zip, push it to your own repository, branch, PR, anything you'd do with a normal repo. Subsequent generations land on a fresh branch so your edits aren't overwritten.
What if I need a small change, not a full regeneration?
File an addendum against the completed package — it produces a focused markdown bundle covering just the change, traceable back to the original. Faster, cheaper, and the original spec stays intact.
Data, privacy & AI providers
Where does my data live?
Azure-hosted in the United States. Interview transcripts and reference documents are encrypted at rest. See the Privacy Policy for the full data-handling breakdown.
Do you store my source code?
No. SpecStep generates documentation from the interview and any reference materials you attach — we don't ingest your codebase. The output is a markdown package that lives in your account until you delete it.
Do you train models on my data? Do the LLM providers?
No. SpecStep doesn't train models on your interview content, reference documents, generated packages, or the personal information in your feedback. We send generation data to third-party AI providers (currently Anthropic and OpenAI) under their no-training-on-API-traffic commercial terms.
Do you use AI on my support or contact messages?
Messages you send through the contact and support flows may be classified, summarized, and routed with AI-assisted tools to get them to the right place faster. A human is responsible for reviewing your message and for any reply — we don't send AI-generated replies without human review.
Can I delete my data?
Yes — from Settings → Delete account. Individual interviews, generations, and packages can also be deleted from the workspace and recovered from the Recycle Bin for 30 days before permanent removal. Detailed audit-log records are kept for 90 days (billing-related records longer, as accounting law requires); see the Privacy Policy for retention details.
Integrations & connectors
Where does the generated package go?
Your choice. Download the zip from the workspace, or pull it via the REST API or MCP. Automated delivery to a GitHub repository as a pull request is on the roadmap.
Do you support other source-control providers?
GitHub is the first provider — connect a repository to a project to track build progress today, with automated pull-request delivery on the roadmap. Azure DevOps, GitLab, and Bitbucket follow. The release notes are the source of truth for what's shipped.
Can I pull reference docs from SharePoint, OneDrive, or Google Drive?
Yes. Connect a SharePoint, OneDrive, or Google Drive folder during the interview and SpecStep reads the contents and feeds them into your spec as reference documents. (Dropbox is coming next; see the Pricing page for what each plan includes.)
What happens to connector files when I disconnect a folder?
Files pulled from a connected folder are saved as reference-document records tied to the interview that used them, so you can reuse them without reconnecting. Disconnecting the folder stops SpecStep from reading it again, but it doesn't delete documents already imported — remove those by deleting the relevant interview or your account. SpecStep never writes to or deletes anything in your cloud storage.
Do you support webhooks?
Yes. Subscribe to generation lifecycle events from Settings → Webhooks. Deliveries are signed, retried with exponential backoff, and you can rotate the secret or send a test payload from the same panel.
Programmable surface (MCP + REST)
What is MCP, and do I need it?
MCP (Model Context Protocol) lets your AI coder call SpecStep tools directly — running interviews, starting generations, fetching packages — from inside Claude Desktop, Cursor, Codex, or GitHub Copilot. You don't need it to use the web app, but if your AI coder is already where the work happens, it removes the context switch.
Do I need an API key for MCP?
No. MCP clients can sign in through the browser — you approve once and the client gets a 90-day token automatically. API keys still work for headless and CI flows; pick whichever fits your setup.
How do I install the MCP server in my AI coder?
The MCP documentation has copy-paste snippets for each client. Most setups are one config-file edit and a client restart.
Account management
I deleted something by accident — can I get it back?
Yes, if it's been less than 30 days. Open the Recycle Bin from the workspace, find the item, and restore it. After 30 days, deleted items are permanently removed.
Can I add someone else to my account?
Multi-user workspaces ship with the Team plan, which is on the way. Until then, each account is single-user; for early access, drop us a line.
Support, feedback & bugs
Use the route that fits what you need — it gets you the fastest answer:
- Something's broken → file a support ticket (or a bug report from inside the app). It lands in the support queue and we reply within one business day.
- Account, billing, or usage help → support ticket.
- Product feedback or a feature request → the Feedback page. Signed-in feedback can attach your account, plan, build, and package or generation context so we can triage it faster.
- Sales, pricing, partnerships, or press → the contact form.
- Security vulnerability → email security@specstep.com (scope and safe-harbor terms on the Security page).