Spec0

Spec0 is a real-time, governed registry for enterprise internal APIs. Its core goal is to provide AI agents with access to the latest, precise OpenAPI specifications, avoiding the need to paste large and easily outdated specification documents directly into the context.

Purpose Spec0 lets MCP-enabled AI agents such as Cursor and Claude Code query a team's private OpenAPI specifications in real time through the API Registry MCP. Each query retrieves only the needed operation (about 300 tokens), rather than injecting an entire specification of about 150k tokens at once, thereby keeping context lean and information always up to date.

Main capabilities

• Intelligent API registry: semantically indexes the API catalog of the entire organization, allowing agents to locate the correct operation among hundreds of APIs with a single query, rather than handling only one specification at a time.
• Three push methods, unified real-time registration: specifications can be pushed directly via CI, Web UI, or CLI; each specification is indexed into operations, components, and metadata, and governed according to team structure.
• Governance and ownership: APIs know which team they belong to, and governance, change alerts, and subscription routing are automatically sent to the correct owners.
• Git-native lifecycle: every push undergoes Spectral validation and breaking change detection, allowing non-compliant specifications to be intercepted before agents see them.
• Organization-level governance: supports custom Spectral rulesets, organization-wide RBAC, audit logs for every MCP query, and SSO.
• Real-time MCP server: specifications update immediately as code is merged, with no manual refresh or re-pasting required.
• Schema-correct Mock: pushing a specification provides a real-time schema-based Mock URL with no infrastructure required; when the specification changes, the Mock changes accordingly. Supports expression rules to simulate real behavior (such as returning 404 when out of stock, returning 402 when the balance is zero, and returning different payloads based on query parameters).
• Compatible with existing MCP clients: Cursor, Claude Code, Windsurf, and any MCP-enabled client can be used.
• Q&A with citations: Ask Spec0 answers questions based on the organization's published specifications and their relationship graph, and links to the cited operations for easy verification.
• Shared Mock URL and CI contract testing: aalways-up-to-date link is available to the entire organization and consumers, and can be integrated into CI to run contract tests on every push, catching backward-incompatible changes at the PR stage.

How to use

1. Register an account: register at app.spec0.io, and the system will immediately provision a private registry and a unique MCP endpoint for the organization.
2. Push specifications from CI: install the CLI, authenticate once, and push private OpenAPI specifications; this can be added to CI to stay up to date after every merge.
3. Configure the agent's MCP client: a single CLI command can configure clients such as Cursor, Claude Code, and Windsurf, and the organization bearer token will be written automatically. After that, you can ask the agent about any API in the catalog.

Installation and command examples (keep the original text): $ npm install -g @spec0/cli $ spec0 auth login $ spec0 push openapi.yaml $ spec0 mcp install --client cursor

Applicable scenarios

• Engineering teams that have internal OpenAPI specifications and want AI agents to query them directly.
• Engineering organizations of about 50-500 people that use Cursor or Claude Code for daily development.
• Teams that need to detect breaking changes at the PR stage and use real-time Mocks early in frontend development.

Limitations and notes

• Official materials do not provide specific pricing information; they only mention that early design partners can receive six months of the Team plan for free (in exchange for two 30-minute conversations).
• Official materials do not provide platform compatibility details (other than MCP clients).
• Official materials do not provide clear effectiveness data or performance metrics.
• Specifications are hosted in the EU and encrypted at rest; the CLI and Mock server are open source (MIT) and can be self-hosted in a VPC or CI.