AnotherAI: let Claude Code build other AI agents. [alpha]

Like you, we fell in love with Claude Code and Cursor. Watching the AI work for a few minutes, vibecoding anything you want using English, feels like magic. It seems like all AI need are better tools to achieve even more.

So what happens when we give Claude Code better tools to build AI agents? We've tried and the early results are promising. We wanted to share an early preview with you.

DEMOs

> Claude: can you build with AnotherAI an agent that can extract calendar events from
> an email? Make sure to try 3 different models from OpenAI, Anthropic and Google,
> and 5 different emails. Tell me what you think.

Watch the magic happen:

Claude Code will:

• Understand how AnotherAI works by using the search_documentation tool
• Pick the models to try using list_models tool
• Use playground tool to compare the models
• Report back the results

And you can see the results in the web app:

> Claude: can you try GPT-5 on 10 random completions from agent:
> extract-calendar-events? what do you think?

Sit back and enjoy the show 🍿

> Claude: can you show me how much $ we're spending daily, by agent?

Claude will write the SQL query to create the view, and you can see the results in the web app:

> Claude: why is this anotherai/completion/01989ada-2185-739f-f5c2-7bb67faaad19 wrong?
> Can you help fix?

Claude will fetch the completions, and fix the issue most of the time!

[!NOTE] This is an early release of AnotherAI. We do not recommend using it in production yet.

How it works

AnotherAI has four main components:

1. An OpenAI compatible API, which is super great for compatibility with any programming language or SDKs, but that supports all the models and logs all the completions.
2. A MCP server that exposes tools designed specifically for Claude Code (and Cursor) to build agents, compare models/prompts (this concept is called experiments), debug issues using the logs, and getting insights from logs (ask any question in English you want).
3. A simple web interface to be able to review the work that the AI did in a nicer UX than a chat/terminal UI (for example comparing different prompts side-by-side), look at the completions logs, and charts that your AI prepared for you.
4. An API for everything else you want to build on top of AnotherAI, giving you a flexibility that we like when vibe-coding (custom evaluations like LLM-as-a-judge, querying the logs database directly, ..).

OpenAI compatible API

Available at http://localhost:8000/v1/chat/completions, this API endpoint accepts any library that already uses OpenAI. To start using the endpoint, simply set the base_url to http://localhost:8000/v1 on most OpenAI or OpenAI compatible SDKs.

> Claude: how many models from how many providers are supported by AnotherAI?

Based on the data from AnotherAI's API, I can see that AnotherAI supports 95 models from 12 different providers: ...

The easiest way to get started is to ask Claude Code to setup AnotherAI on your agent. For example:

Claude: can you setup AnotherAI on this agent @my-agent.py?

MCP server

We use the http transport to expose the MCP server at https://api.anotherai.dev/mcp. See below for instructions on how to install it in Claude Code and Cursor.

We have curated a list of tools that we think are useful for building AI agents, available at [backend/protocol/api/_mcp.py](blob/main/b

... View full README on GitHub