@texoport/effect-ai-claude

This lets an Effect AI program talk to the claude CLI that is already logged in on your machine. No Anthropic API key. No second account to wire up. It is a useful fit for local tools, personal projects, and anything else where Claude Code is already part of the setup.

It expects Claude Code Pro or Max, claude on PATH, and a working CLI login. Check that first:

claude auth status

Install

pnpm add @texoport/effect-ai-claude [email protected] @effect/[email protected]

Ask Claude something

model gives Effect AI a normal LanguageModel layer. The rest of the program uses the same API as any other Effect provider.

import { NodeRuntime, NodeServices } from "@effect/platform-node"
import * as ClaudeLanguageModel from "@texoport/effect-ai-claude"
import { Effect } from "effect"
import { LanguageModel } from "effect/unstable/ai"

const program = Effect.gen(function* () {
  const response = yield* LanguageModel.generateText({
    prompt: "Explain what a semaphore does in two sentences."
  })

  console.log(response.text)
}).pipe(
  Effect.provide(ClaudeLanguageModel.model("sonnet")),
  Effect.provide(NodeServices.layer)
)

NodeRuntime.runMain(program)

Tools

Tool calls work through the normal Effect Toolkit API. This package starts a small local MCP bridge for the turn, so your tool handlers stay in the same Node process as your application. Claude sees them as MCP tools through claude -p.

Configuration

ClaudeLanguageModel.model("sonnet", {
  timeout: "2 minutes",
  debug: true
})

bin points at a different Claude executable. cwd changes the CLI working directory. extraArgs passes flags through to claude. Calls time out after five minutes unless you set timeout. Set debug when you want spawn timing and Claude's stderr in your terminal.

License

MIT