LongCat 2.0 · configuration
The official LongCat integration guide configures a named client and keeps the first run observable and reversible.
LongCat-compatible client, endpoint, model ID, and an isolated test workspace.
Configure the endpoint and model alias, run a read-only repository task, then record the tool calls and diff.
Produce a small code change with an auditable client configuration.
The request reaches the intended model; the diff stays in scope; tests and tool returns are saved.
If the client falls back to another model, inspect the request and alias before editing code.
The official integration guide documents configuration, not coding quality or production reliability.
By rewriting the environment variables in ~/.claude/settings.json, you can point all four Claude Code model slots (primary model / small fast model / Sonnet / Opus) to LongCat-2.0's Anthropic-compatible endpoint, allowing Claude Code to use LongCat directly for repository-level coding tasks.
Suitable tasks: Use Claude Code as an Agent shell with LongCat-2.0 as the backend for code understanding, repository-level changes, and automated task execution; compare LongCat with the Claude family in real coding workflows
Unsuitable tasks: Scenarios that require Claude-specific functionality (such as particular toolchain behavior built into Claude Code) and do not accept behavioral differences
Applicable model version: LongCat-2.0 (Anthropic-format endpoint https://api.longcat.chat/anthropic)
Applicable clients, Agents, or APIs: Claude Code CLI (the official Anthropic client)
Recommended reasoning tier and parameters: The official example sets CLAUDE_CODE_MAX_OUTPUT_TOKENS to 131072 (128K); CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 disables nonessential traffic
~/.claude/settings.json{
"env": {
"ANTHROPIC_AUTH_TOKEN": "your_longcat_api_key",
"ANTHROPIC_BASE_URL": "https://api.longcat.chat/anthropic",
"ANTHROPIC_MODEL": "LongCat-2.0",
"ANTHROPIC_SMALL_FAST_MODEL": "LongCat-2.0",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "LongCat-2.0",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "LongCat-2.0",
"CLAUDE_CODE_MAX_OUTPUT_TOKENS": "131072",
"CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC": 1
},
"permissions": {
"allow": [],
"deny": []
}
}claude
# Enter the test message: "Hello, please introduce yourself"
# Receiving a reply from the LongCat model means the configuration succeededThe settings.json contents were transcribed verbatim from the official documentation (collected on 2026-08-18).
The official model card and OpenRouter page both list "Claude Code" as one of the Harnesses deeply adapted for LongCat-2.0; OpenRouter Apps data shows Claude Code as LongCat-2.0's second-largest traffic source (3.31B tokens, behind Hermes Agent's 7.77B), supporting that this configuration path is used at significant scale in real workflows.
Note: ANTHROPIC_AUTH_TOKEN must contain the LongCat API key, not an Anthropic key; ANTHROPIC_BASE_URL must point to LongCat's Anthropic-compatible endpoint.
This configuration replaces all of Claude Code's model slots with LongCat-2.0, so Claude and LongCat cannot be mixed within the same session; switch back by editing the configuration.
The Anthropic-compatible endpoint is implemented by a third party (Meituan), so compatibility for tool calls and other behavior is subject to the official documentation; OpenRouter testing found a tool-call error rate of approximately 0.90% (see review document 03).
The official documentation also provides equivalent configuration pages for Kilo Code, Codex, ZCode, Cline, Cherry Studio, CodeBuddy, WorkBuddy, CC Switch, and other tools (/platform/docs/<tool name>); use the same method when needed.
LongCat official API documentation site (longcat.chat) · Source date: 2026-06-30 · Edited: 2026-09-20
Read the original sourceLongCat 2.0
Run this guide in the environment listed above. Downloading does not transfer the template or establish model availability for your account.