Loading SpecStep…
From idea to implementation-ready spec

Turn messy product ideas into build-ready specs.

SpecStep interviews you, finds the gaps, and brings in specialist reviewers to produce the requirements, architecture, risks, and acceptance criteria your AI coding tools and team can actually build from — without the rework.

Try free →See how it works
Start from an idea or your existing docs 24 specialists on call Built for your AI coding tools
In preview
The problem

Why AI builds drift

AI coding tools are brilliant at the next ten lines and unreliable at the next ten thousand. The cause is almost never the model — it’s the missing context.

Incomplete context, confident output

Hand an AI coder a vague prompt and it fills the gaps with guesses — plausible, wrong, and expensive to unwind once they’re baked into the code.

Rework instead of progress

Every missing decision becomes a late surprise: re-architecture, rewrites, and sprints that quietly slip while everyone re-litigates what “done” means.

Fragile handoffs

Agents, contractors, and teammates each interpret the idea a little differently — so the pieces don’t line up when they come together.

What you get

From a one-line ask to a build-ready plan

The same messy request, turned into the decisions an AI coder and an engineering team can actually build from.

Before

“Build me a customer portal.”

One sentence. A dozen unanswered questions.

After SpecStep
  • Requirements
  • User flows
  • Auth & access decisions
  • Data model
  • Risks & edge cases
  • Acceptance criteria
  • Test strategy
  • Implementation order
01

Idea

Open a chat with Otto. Talk like a human — no forms, no required fields, no terminology gates.

02

Conversation

Otto asks what matters and pulls in specialists when your project hits their lane. You see who joined and why.

03

Spec

The team that talked to you writes the package. Vision, requirements, ADRs, traceability — ready for your AI coder.

How it works

Otto runs the room. He brings in specialists when your project needs them.

Otto asks, listens, and pulls the right agent in at the right moment. The agents who participate in the conversation will also help write your package, and pull in additional expertise when needed.

Built for your team

Who it’s for

Whoever owns the outcome gets the same source of truth to build from.

Founders

Ship the right thing the first time — without a full-time PM in the room.

Product leaders

Turn strategy and stakeholder input into specs engineering actually trusts.

Engineering leaders

Give every AI coder and contractor one source of truth to build from.

Agencies & consultants

Scope, price, and deliver client builds with far less ambiguity.

Meet the team

24 specialists. One Otto-led conversation.

Otto runs every interview. The other 23 are pulled in when your project hits their lane — and then they stick around to write the parts of the package they own.

No. 01

Otto

Interviewer

Otto runs the conversation. He asks focused questions to draw out what your project actually needs, and pulls in specialists when expertise outside his lane would help.

Signature accent
No. 23

Marc

Business Analyst

Reads the industry. Names the patterns the spec builds against.

Signature accent
No. 02

Stax

Stack Analyst

Stax picks the stack. Languages, frameworks, libraries — and the build-vs-buy calls for the messy concerns like auth, queueing, observability, secrets, and CI/CD.

Signature accent
No. 03

Alan

Architect

Alan drafts the system design. Vision, requirements, ADRs, the phase plan, the testing strategy. He turns the conversation into structure your AI coder can follow.

Signature accent
No. 04

Lyra

Designer

Lyra owns the visuals. She validates UI proposals against your design references and writes the visual-inspection protocol the build agent runs before it ships.

Signature accent
No. 05

Vetta

Reviewer

Vetta catches what slipped past. Missing sections, contradictions between sections, ambiguous acceptance criteria — the things that block shipping if they reach code.

Signature accent
No. 06

Iris

Fresh Eyes

Iris reviews the package on a different LLM provider than the rest of the team. She catches model-specific blindspots — the issues groupthink lets through.

Signature accent
No. 07

Sterling

Editor

Sterling assembles the final package. The markdown, the tool-specific mirrors (Claude Code / Cursor / Cline), the manifest. No new reasoning — just the right shape, every time.

Signature accent
No. 08

Linus

Code Reviewer

Finds the bugs. Skeptical, rigorous, never hand-wavy.

Signature accent
No. 09

Vera

Test Automation

Writes the test suite. Coverage, edge cases.

Signature accent
No. 10

Quill

Copy Editor

Polishes copy. Voice, rhythm, the right word.

Signature accent
No. 11

Hush

Privacy Attorney

Privacy and data compliance. GDPR, CCPA, the lot.

Signature accent
No. 12

Mercer

Commercial Attorney

Contracts, vendor agreements, commercial terms.

Signature accent
No. 13

Ada

Internet & Tech Attorney

Internet law, IP, platform terms, technology contracts.

Signature accent
No. 14

Argus

Security Expert

Threat modeling, hardening, incident response.

Signature accent
No. 15

Atlas

Reliability Engineer

Carries the load. SLOs, alerts, the on-call story.

Signature accent
No. 16

Codd

Data Architect

Designs the data. Schemas, indexes, migrations that don't lock.

Signature accent
No. 17

Halo

Accessibility Expert

Accessibility-first. WCAG, screen readers, real keyboards.

Signature accent
No. 18

Polo

Localization Expert

Speaks the languages. Locales, plurals, RTL, translation flow.

Signature accent
No. 19

Merlin

Prompt Engineer

Tunes the AI. Models, prompts, evals, the cost ceiling.

Signature accent
No. 20

Reg

Compliance Specialist

SOC 2, HIPAA, PCI. Framework controls, audit evidence.

Signature accent
No. 21

Tally

Cost Engineer

Counts the cost. Build hours, run-rate, the dollar truth.

Signature accent
No. 22

Hazard

Risk Officer

Names the risks. Schedule, vendor, regulatory, the AI-coder traps.

Signature accent
No. 24

Trip

UX Researcher / Workflow Analyst

Walks every flow. Edge cases, empty states, the offramps.

Signature accent
MCP-native

Your AI coder reads it directly.

Every spec ships with a programmable surface. Drive the full pipeline from your agent or CI: start an interview, kick off a generation, fetch a package, register a webhook. Same sf_* API key on every surface.

70 MCP tools

Full coverage of interviews, intake artifacts, generations, packages, change requests, webhooks. Your agent can drive end-to-end without hand-crafting HTTP.

Browse the tool catalog →

REST + OpenAPI

Conventional JSON API at specstep.com/v1/. Generate SDKs from /v1/openapi.json; wire it into CI; write integration scripts in any language.

Read the REST walkthrough →

Webhooks

Subscribe to generation.completed, generation.failed, and the rest. HMAC-SHA256 signatures, bounded retry, dedup-friendly delivery ids.

See webhook delivery →
Picked up natively by:
Claude CodeCLI agent — native shape
CursorWorkspace rules + indexing
CopilotCustom instructions
CodexCLI + chat surfaces
External connectors

Connect the folder you've already been working in.

You probably already have a folder. A brief, some sketches, a half-written requirements doc. Connect it once and SpecStep reads it before the first question — summarizes what it sees and asks you to confirm. The interview starts where your thinking left off, not from scratch.

OneDrive
SharePoint
Google Drive
Dropbox Coming soon
Pro+ Free accounts can connect a folder and watch the interviewer summarize it. Generating the spec from connector data needs Pro or Team.
What you get

A complete documentation package — purpose-built for the way AI coders read context.

Every specification package produces a structured document shaped for the way AI coders read context. Your spec becomes the source of truth your build agent can't drift from.

ledger-cli/
├── 00-vision.md           # Otto + Alan
├── 01-requirements/
│   ├── R-AUTH-014.md
│   ├── R-LEDGER-021.md     # Vetta reviewed
│   └── ...
├── 02-architecture/
│   ├── ADR-0007-postgres-vs-sqlite.md
│   ├── ADR-0012-cli-args-vs-config.md
│   └── system-overview.md  # Alan
├── 03-design/            # Lyra (because UI was in scope)
│   ├── visual-inspection-protocol.md
│   └── mockups/          # HTML/CSS mockups
│       ├── README.md
│       └── dashboard.html
├── 04-security/          # Argus (added mid-interview)
│   └── threat-model.md
├── 05-phase-plan.md
├── 06-traceability.csv
├── 07-session-protocol.md # Sterling
└── manifest.yaml

Vision & requirements

Numbered, traceable, AI-coder-shaped. Each requirement links forward to ADRs and back to a moment in your conversation.
Otto + Alan

Architecture decisions

Real ADRs with status, consequences, and the trade-offs the team weighed. Not "here's a stack" — here's why.
Alan

Domain-specific sections

Got UI? Lyra writes the visual protocol AND drafts HTML mockups for your key surfaces. Building with Claude or another AI? Merlin tunes prompts. Storing PII? Hush flags it. Multi-language? Polo plans i18n. Public auth surface? Argus models the threats. The package only grows what your project needs.
Lyra, Merlin, Polo, Argus & co.

Phase plan + session protocol

Sterling assembles the build order and the session rules — so your AI coder knows what to read first, what to verify between turns, and where to stop.
Sterling
Sign in

Ready to spec your project?

Talk to Otto. Watch the team form around your idea. Hand the package to your AI coder and ship.