# Entalpa

> Entalpa is a spec-driven development platform for AI coding agents: a persistent, MCP-native specification layer (stakeholders → business needs → user stories → requirements → design specs) with atomic locking and open questions surfaced before the build, for people building with Claude Code, Cursor and Codex. It turns an ambiguous product idea into structured, traceable requirements and serves them to the agent over MCP, so the agent builds what was meant without breaking what exists.

Key facts:

- Category: spec-driven development (SDD). Unlike file-based SDD workflows (GitHub Spec Kit, OpenSpec, Kiro's per-feature specs), the spec is a store, not a markdown file: it persists across sessions, and a locked requirement is excluded from what the model is allowed to regenerate.
- Problem it solves: architectural drift in AI-assisted ("vibe") coding. Agents guess implicit business logic, break existing features, rewrite the spec they were given, and burn tokens re-deriving context every session, because no durable, governed spec exists upstream of the code.
- Approach: a living, traceable specification the agent has to follow, served over MCP. End-to-end traceability from stakeholder intent to design. Entalpa does not replace coding agents; it stabilizes them. It imports GitHub Spec Kit (`specs/`) and OpenSpec (`openspec/specs/`) repositories into one current, viewable, auditable and traceable source of truth, and the team can keep using those workflows for the next change.
- Audience: solo founders and small teams building real products with AI agents; product people who need a spec their engineers' agents will actually read.
- Access: web app at https://app.entalpa.com with a free tier (one credit a week, up to 2; no credit card required) and monthly plans from $9 (https://entalpa.com/en/pricing); reading and writing a project over MCP never spends credits; no lock-in, agent-agnostic (any MCP client). Export to ReqIF.

## Pages

- [Home](https://entalpa.com/en): what Entalpa does, the failure modes of AI coding agents it fixes, supported agents, FAQ
- [For product teams](https://entalpa.com/en/for-product-teams): the same platform pitched to product managers and founders
- [Spec-driven development](https://entalpa.com/en/spec-driven-development): Entalpa as a spec-driven development tool for people already working that way: a spec drafted from a paragraph instead of written by hand, one current spec of the whole system instead of a folder per feature, locked requirements the model is never handed to rewrite, open questions before the code, and imports from Spec Kit and OpenSpec
- [Docs](https://entalpa.com/en/docs): what Entalpa's ideas mean and how to do the main tasks: the specification pipeline (stakeholders, user stories, requirements, design) and how people and coding agents build from it
  - [Getting started](https://entalpa.com/en/docs/getting-started): six steps from a plain-language idea to a spec your team builds from
  - [Requirements and their lifecycle](https://entalpa.com/en/docs/requirements): what a requirement holds, its nine states from draft to verified, the status bar over the requirements, how a story's state follows from its requirements, and the standards and research its fields and states follow (ISO/IEC/IEEE 29148, the INCOSE Guide to Writing Requirements, Wiegers and Beatty's requirement statuses, Gotel and Finkelstein on traceability)
  - [The Spec score](https://entalpa.com/en/docs/spec-score): one number from 0 to 100 for how ready a spec is to build from: its six areas and weights, the caps, when it moves, what to fix first, and the standards and research each check comes from (ISO/IEC/IEEE 29148, ISO/IEC 25010, the INCOSE Guide to Writing Requirements, the Quality User Story framework, INVEST)
  - [Subsystems and propagation](https://entalpa.com/en/docs/subsystems): splitting a project into subsystems, how a subsystem's requirements trace to the system requirements, drafting them with AI, and the systems-engineering sources behind it (ISO/IEC/IEEE 15288 architecture definition, the INCOSE Needs and Requirements Manual on allocation and flow-down, coupling and cohesion, Twin Peaks, Parnas on information hiding)
  - [Interface contracts](https://entalpa.com/en/docs/interface-contracts): how one subsystem depends on another (data, trigger, state), the read-only interface requirement each contract becomes, what the Spec score checks, and the sources behind it (interface management in the INCOSE and NASA systems engineering handbooks, ISO/IEC/IEEE 29148 interface requirements, Meyer's design by contract)
  - [Suspect links](https://entalpa.com/en/docs/suspect-links): requirements whose story changed after they were written, where they are marked, the three ways to clear the mark, and the traceability and change-management sources behind it (Gotel and Finkelstein, event-based traceability, ISO/IEC/IEEE 29148, ISO 26262-8, Automotive SPICE)
  - [Updating an import, and exporting](https://entalpa.com/en/docs/import-export): what an OpenSpec or Spec Kit import brings in, updating a Spec Kit import with new features, and every export (OpenSpec, Spec Kit, ReqIF, CSV, Markdown, portable JSON/YAML, PRD)
  - [The project file: JSON and YAML](https://entalpa.com/en/docs/project-file): the portable project file in JSON or YAML, which to choose (YAML to keep the spec in a repository and review its diffs), what it keeps and leaves out, and how importing it makes a new project
  - [Generating and regenerating](https://entalpa.com/en/docs/generating): what Entalpa's AI drafts and when, what is added at once and what waits for review, what a lock protects, and why a person reviews what AI drafts (ISO/IEC/IEEE 15288 needs and requirements, research on language models in requirements engineering)
  - [The chat and its suggestions](https://entalpa.com/en/docs/chat): the project's chat, what it asks to confirm first, and how its proposals reach the review
  - [Versions and undo](https://entalpa.com/en/docs/versions): saving, comparing and restoring versions, undoing changes, what a comparison is for (a coding agent implementing only what changed, a reviewer, a process that must record its changes), and baselines and change control in the standards (ISO/IEC/IEEE 15288, ISO 13485, IEC 62304, ISO 26262-8, Automotive SPICE)
  - [Sharing a project](https://entalpa.com/en/docs/sharing): a read-only public link to one version, members, and previews of an import tried before signing up
  - [What an agent can do](https://entalpa.com/en/docs/agents): how a coding agent connects to Entalpa's MCP server and signs in as the user (OAuth, the scopes it asks for), what it can read, change and delete, which tools use AI and spend credits, what locks do and do not stop, and what stays the user's call
  - [MCP tool reference](https://entalpa.com/en/docs/mcp-tools): every tool Entalpa's MCP server gives an agent, by area, with whether it reads, writes or can delete and whether it uses AI; generated from the server's own tool list
  - [Skills](https://entalpa.com/en/docs/skills): the four Entalpa skills for coding agents (entalpa-implement, entalpa-prd, entalpa-openspec, entalpa-speckit), what each is for, and how to install them
  - [Troubleshooting](https://entalpa.com/en/docs/troubleshooting): what to check when an agent cannot sign in, shows no tools or no projects, or does not use a skill
- [Connect your agent](https://entalpa.com/en/connect): the Entalpa MCP server, with a setup page for each client (Claude Code, Cursor, Codex, Windsurf, VS Code + Copilot, Gemini CLI, Claude Desktop, Cline, Roo Code, Zed, Antigravity, Kiro, Warp, OpenCode, Amp, Continue)
- [Import from OpenSpec](https://entalpa.com/en/openspec): paste a public GitHub repository with an `openspec/` folder, or upload a private one as a zip or a folder, and read it as one project (stakeholders, stories, requirements, open questions) before signing up
- [Import from Spec Kit](https://entalpa.com/en/spec-kit): import a GitHub Spec Kit repository's `specs/` folder (public on GitHub, or a private one as a zip or a folder) as one backlog, with statuses from `tasks.md`
- [Spec-driven development tools compared](https://entalpa.com/en/vs): one table across Spec Kit, Kiro, OpenSpec, BMAD, Tessl and Entalpa by what a single source of truth needs, then an honest page on each, including CLAUDE.md, Context7, memory MCP servers, ChatPRD, Jama and DOORS, each saying where the other tool wins
- [Guides](https://entalpa.com/en/guides): long-form pieces built from what developers said about their own projects and from incidents in ours. What spec-driven development is and what happened to it, spec-driven development with Claude Code, the vibe-coding wall, what your agent breaks, reviewing the spec not the diff, why agents guess requirements, and a PRD template written for coding agents
- [Agent failure modes, measured](https://entalpa.com/en/fix): one page per failure mode, each with a measurement we ran on a real repository and the method printed on the page
- [Pricing](https://entalpa.com/en/pricing): Free (one credit a week, up to 2), Plus ($9 a month, 20 credits, up to 60), Pro ($49 a month, 200 credits, up to 1,000), and one-time top-ups at $1 a credit
- [App](https://app.entalpa.com): the Entalpa web app; sign up and start free, no credit card required

## Contact

- Email: contact@entalpa.com
- GitHub: https://github.com/entalpa
- LinkedIn: https://www.linkedin.com/company/entalpa
- YouTube: https://www.youtube.com/@entalpa_ai
- App: https://app.entalpa.com
- Company: Entalpa

## Optional

- [Privacy policy](https://entalpa.com/en/privacy-policy)
- [Terms of service](https://entalpa.com/en/terms-of-service)