AI agents shouldn't need browser automation
Mobile and web apps talk to private APIs all day.
Spectral gives that same access to your AI agents.
$ curl -LsSf https://getspectral.sh/install.sh | bash or install with pip / uv
How it works
Capture
Chrome extension or MITM proxy records traffic and GUI context while you browse
Analyze
LLM correlates your actions with API calls and generates MCP tools
Authenticate
Auto-detects auth flows and generates a login script
Use
AI agents call the real API directly via MCP server
Demo
Capture traffic, analyze with LLM, login and use via MCP
Why Spectral
Works everywhere
Websites, Android, desktop, CLI
Understands meaning
Correlates UI actions with API calls
Self-healing tools
Auto-corrects failed tools at runtime
Bring your own LLM
Anthropic, OpenAI, OpenRouter, Ollama — only needed at build time
Faster than browser automation
No headless browser, no selectors
Also generates API specs
OpenAPI 3.1 and GraphQL SDL
Safer than letting your agent browse
Instead of giving your AI agent a browser or raw curl access, Spectral provides a controlled layer between the agent and the APIs it uses.
Credentials stay hidden
Your agent never sees API keys or tokens — Spectral injects authentication at runtime, outside the LLM context.
Scoped access
Expose only the tools you choose — read-only, specific endpoints, or a curated subset. Your agent can't wander beyond what you allow.
Declarative, not executable
MCP tools are HTTP request templates, not scripts. No arbitrary code runs — just structured API calls your agent fills in.
Community-reviewed catalog
Every community tool goes through a pull request review before joining the catalog. Everything is public and auditable.