Frigade Reverse-Engineers Any Web App Into Agent Tools
Frigade hit the Hacker News front page today with a sharp idea: instead of waiting for every SaaS product to ship an official MCP server, reverse-engineer one out of the app itself. A browser agent runs inside your authenticated web app, watches the API calls the frontend actually makes, and generates recipes β endpoints, auth, response schemas, descriptions. The output is effectively a self-updating MCP server with no manual integration code.
This is a third way in the agent-access debate. Official APIs and MCP servers do not exist for most of the software people actually use, and computer-use agents that click through UIs are slow, brittle and non-deterministic. Frigade skips both: it learns the private API the app already runs on. Demos are live for Jira, Spotify, Hacker News and Airbnb, auth complexity like JWTs and cookies is handled automatically, and calls execute client-side so nothing routes through their servers.
The honest caveats: it is not open source (the team says they are considering it), GraphQL apps are the hard case by their own admission, and private APIs can change under you without notice β the self-updating part is doing a lot of load-bearing work.
Still, the direction feels right. Coasty attacked legacy software with vision and clicks last week; Frigade attacks it at the network layer. The agent-access problem is being squeezed from both ends, and the software vendors who never planned to ship an API are running out of places to hide.
Demo: https://demo.frigade.com/ How it works: https://frigade.com/how-it-works
← Back to all articles
This is a third way in the agent-access debate. Official APIs and MCP servers do not exist for most of the software people actually use, and computer-use agents that click through UIs are slow, brittle and non-deterministic. Frigade skips both: it learns the private API the app already runs on. Demos are live for Jira, Spotify, Hacker News and Airbnb, auth complexity like JWTs and cookies is handled automatically, and calls execute client-side so nothing routes through their servers.
The honest caveats: it is not open source (the team says they are considering it), GraphQL apps are the hard case by their own admission, and private APIs can change under you without notice β the self-updating part is doing a lot of load-bearing work.
Still, the direction feels right. Coasty attacked legacy software with vision and clicks last week; Frigade attacks it at the network layer. The agent-access problem is being squeezed from both ends, and the software vendors who never planned to ship an API are running out of places to hide.
Demo: https://demo.frigade.com/ How it works: https://frigade.com/how-it-works
Comments