AI Coding Tools That Support a Custom API (2026)
As of 2026-10-08, checked against each tool's official docs, all 16 AI coding tools on this page let you bring your own API key and change the base URL: 14 of them officially support the Anthropic protocol, so they can run Claude; Codex CLI speaks only OpenAI Responses, so it cannot call Claude on QCode; for Cursor, no setting to change the Anthropic row's base URL was found in its official docs (they describe an override switch only for the OpenAI row), so this list does not count it as Claude-capable. Each row below lists the protocol, where to configure it, whether Claude works, and the QCode setup page.
Updated 2026-10-08
Four numbers first
Tools checked
Each one has a setup page on docs.qcode.cc, and its official docs contain a verbatim line about a custom endpoint or BYOK; checked on 2026-10-08.
Can use Claude
The official docs state Anthropic protocol support, or that the Anthropic provider accepts a custom base URL.
Claude unverified or unavailable
Codex CLI speaks only OpenAI Responses, and the QCode docs say Claude does not work there; for Cursor, as of 2026-10-08 no setting to change the Anthropic row's base URL was found in its official docs. For Claude, use a tool that supports the Anthropic protocol, or Claude Code.
Two base URLs
Anthropic protocol: https://api.qcode.cc/api. OpenAI-compatible: https://api.qcode.cc/openai/v1. A few tools expect another form; use the value on their setup page.
The list, part 1: editors and IDE extensions
As of 2026-10-08, all 16 popular AI coding tools below accept a custom API; 14 can reach Claude over the Anthropic protocol, while for Cursor and Codex CLI only an OpenAI-protocol custom address turned up in the official docs. Editors and IDE extensions come first; each row gives the protocol, where to configure it (setting names as in the official docs), whether Claude works, and the QCode setup page. ① GitHub Copilot: three API types, Chat Completions, Responses and Anthropic Messages; in VS Code add a Custom Endpoint and its settings go into chatLanguageModels.json, while Copilot CLI uses the environment variables COPILOT_PROVIDER_TYPE=anthropic and COPILOT_PROVIDER_BASE_URL; Claude works, and in VS Code the url takes the full address this tool expects, https://api.qcode.cc/api/v1/messages, while the CLI takes https://api.qcode.cc/api; setup page docs.qcode.cc/docs/ide/github-copilot; checked 2026-10-08, source: VS Code docs and GitHub Docs. ② Cline: Anthropic and OpenAI Compatible; pick Anthropic as the API Provider, then tick Use custom base URL; Claude works with https://api.qcode.cc/api; setup page docs.qcode.cc/docs/ide/cline; checked 2026-10-08, source: Cline docs. ③ Roo Code: the same Anthropic plus Use custom base URL as Cline, and OpenAI Compatible; Claude works with https://api.qcode.cc/api; the GitHub repository is archived (read-only); setup page docs.qcode.cc/docs/ide/roo-code; checked 2026-10-08, source: Roo Code docs and the GitHub repository record. ④ Kilo Code: in Settings open the Providers tab, scroll to the bottom and click Custom provider, then choose Anthropic Messages or OpenAI Compatible as the Provider API; Claude works with https://api.qcode.cc/api; setup page docs.qcode.cc/docs/ide/kilo-code; checked 2026-10-08, source: Kilo docs. ⑤ Zed: under language_models in settings.json there are anthropic_compatible and openai_compatible providers, with the address in api_url; Claude works with https://api.qcode.cc/api; the docs warn not to put API keys in settings.json; setup page docs.qcode.cc/docs/ide/zed; checked 2026-10-08, source: Zed docs. ⑥ Trae (now called TraeCode in its docs): add a custom model under Settings > Models, with Anthropic Messages or OpenAI Chat Completions as the format and a separate full-URL switch; Claude works with https://api.qcode.cc/api; setup page docs.qcode.cc/docs/ide/trae; checked 2026-10-08, source: Trae docs. ⑦ JetBrains IDEs (Claude Code plugin): the plugin runs the claude command in the IDE terminal and does not bundle its own CLI, so it uses Claude Code's ANTHROPIC_BASE_URL; Claude works with https://api.qcode.cc/api; setup page docs.qcode.cc/docs/ide/jetbrains; checked 2026-10-08, source: Claude Code docs.
The list, part 2: command-line agents and desktop clients
⑧ Claude Code (CLI and VS Code extension): Anthropic protocol; the environment variable ANTHROPIC_BASE_URL, which the docs describe as pointing the API endpoint at a proxy or gateway, and the VS Code extension shares ~/.claude/settings.json with the CLI; Claude works with https://api.qcode.cc/api (no trailing slash); setup pages docs.qcode.cc/docs/getting-started/environment and docs.qcode.cc/docs/ide/vscode; checked 2026-10-08, source: Claude Code docs. ⑨ Claude Desktop: connects only to a gateway that implements the Anthropic Messages API; open Developer → Configure Third-Party Inference and set Inference provider to Gateway; Claude works with https://api.qcode.cc/api; setup page docs.qcode.cc/docs/ide/claude-desktop; checked 2026-10-08, source: Claude Desktop third-party inference docs. ⑩ Cherry Studio: a custom provider can point at OpenAI-, Anthropic- and other compatible endpoints; Settings → Model Services → add a provider; Claude works with the Anthropic type and https://api.qcode.cc/api, while the OpenAI type, in this tool's own form, takes only https://api.qcode.cc/openai and the app appends the path; setup page docs.qcode.cc/docs/ide/cherry-studio; checked 2026-10-08, source: Cherry Studio docs. ⑪ OpenCode: any provider's address can be changed with baseURL, and any OpenAI-compatible service can be added; global config ~/.config/opencode/opencode.json; Claude works with this tool's form, https://api.qcode.cc/claude/v1; the QCode docs note that OpenCode needs direct API request access enabled first; setup page docs.qcode.cc/docs/ide/opencode; checked 2026-10-08, source: OpenCode docs. ⑫ Aider: the docs say it can connect to any OpenAI-compatible endpoint (OPENAI_API_BASE), with the Claude key in ANTHROPIC_API_KEY; Claude works, and the QCode docs use an anthropic/ prefix on the model name with ANTHROPIC_API_BASE set to https://api.qcode.cc/api, a variable documented by LiteLLM, the library Aider uses; setup page docs.qcode.cc/docs/ide/aider; checked 2026-10-08, source: Aider and LiteLLM docs. ⑬ Crush: custom providers come in OpenAI-compatible and Anthropic-compatible types; global config ~/.config/crush/crushrc (the docs call crush.json the original, now deprecated format, which the setup page still uses in its example); Claude works with https://api.qcode.cc/api; setup page docs.qcode.cc/docs/ide/crush; checked 2026-10-08, source: Crush repository docs. ⑭ Droid (Factory): BYOK goes in the customModels array of ~/.factory/settings.json, and provider anthropic covers the official API or compatible proxies; Claude works with https://api.qcode.cc/api; the docs say API keys stay local and are not uploaded to Factory servers; setup page docs.qcode.cc/docs/ide/droid; checked 2026-10-08, source: Factory docs.
Milestones for BYOK entry points
GitHub changelog: Copilot CLI gains BYOK; set a few environment variables before launch to use Azure OpenAI, Anthropic or any OpenAI-compatible endpoint.
VS Code 1.122 ships with the Custom Endpoint provider in Stable; from the same release BYOK works for chat, tools and MCP without a GitHub sign-in.
All 16 tools on this page were checked row by row against their official docs and the docs.qcode.cc setup pages.
The two documented for OpenAI only, and what could not be verified
Verbatim in the official docs: the two not counted for Claude
As of 2026-10-08, only an OpenAI-protocol custom address turned up in the official docs of the two tools below, so this list does not count them as Claude-capable; for Claude, use a tool that supports the Anthropic protocol, or Claude Code. ⑮ Cursor: the help center's BYOK steps are to open Cursor Settings > Models, find the provider and paste the key; the official release notes give the OpenAI row a Base URL override switch, and as of 2026-10-08 no setting to change the Anthropic row's base URL was found in Cursor's official docs; custom keys apply to chat models only, and Tab completion keeps using Cursor's built-in models; the OpenAI row can run GPT and the Chinese model families with the base URL https://api.qcode.cc/openai/v1, but not Claude; for Claude, run Claude Code in Cursor's integrated terminal (ANTHROPIC_BASE_URL set to https://api.qcode.cc/api), where the two paths authenticate separately but share quota when they use the same key; setup page docs.qcode.cc/docs/ide/cursor; checked 2026-10-08, source: Cursor help center and release notes. ⑯ Codex CLI: the docs say a model provider defines the base URL, wire API and authentication, and responses is the only supported wire_api value; config file ~/.codex/config.toml, with base_url = https://api.qcode.cc/openai per the QCode docs; GPT models only, and the QCode docs state that neither Claude nor the Chinese model families work here; setup page docs.qcode.cc/docs/ide/codex; checked 2026-10-08, source: OpenAI Codex configuration docs.
Not verified, or differing from the QCode docs
As of 2026-10-08, a few rows follow the more conservative source: ① Cursor: the QCode Cursor page says the Anthropic row can also override its base URL, citing a Cursor staff reply on the forum, but as of 2026-10-08 no setting to change the Anthropic row's base URL was found in Cursor's official docs, so this list does not count Cursor as Claude-capable; ② Codex CLI: the QCode tool compatibility overview lists OpenAI Responses and Chat, while the official configuration reference allows only responses for wire_api, so this list says Responses only; ③ Aider: the Claude address variable ANTHROPIC_API_BASE appears only in the LiteLLM docs, not in Aider's own; ④ Trae: the QCode setup page calls its values a reference that was not tested inside Trae, and Kilo Code's official base URL example includes /v1 while the QCode page stops at /api. Send one test request after setting up any of these. Also, Roo Code's GitHub repository is archived, so no further official fixes are expected.
Anthropic-protocol tools vs tools documented for OpenAI only
Anthropic protocol supported (14 tools)
Use the base URL https://api.qcode.cc/api; Claude and the Chinese model families (GLM, Kimi, DeepSeek, Qwen) all work there, GPT models do not. If you want Claude for coding, pick from this group.
OpenAI only in the official docs (Cursor, Codex CLI)
Cursor takes https://api.qcode.cc/openai/v1 and runs GPT and the Chinese model families; Codex CLI takes https://api.qcode.cc/openai per the QCode docs and runs GPT only. Putting a claude- model on the OpenAI endpoint returns model_not_available_on_endpoint, and that check runs before authentication. For Claude, use a tool that supports the Anthropic protocol, or Claude Code.
Three steps to plug a key into any of them
① Create a key in the QCode console (the docs examples start with cr_). ② Find your tool in the list above: for the Anthropic protocol set the base URL to https://api.qcode.cc/api (no /v1, no trailing slash); for OpenAI-compatible use https://api.qcode.cc/openai/v1; a few tools expect another form, so copy the value from their setup page: GitHub Copilot in VS Code takes the full address https://api.qcode.cc/api/v1/messages, OpenCode's Anthropic form is https://api.qcode.cc/claude/v1, and the OpenAI type in Cherry Studio as well as Codex CLI take only https://api.qcode.cc/openai. ③ Enter a model ID from /models (for example claude-sonnet-5-5, gpt-6.1-sol or deepseek-v4.1-flash) and send one test request: with your key, a request to https://api.qcode.cc/openai/v1/models that returns a JSON list means the address and key are both right; if you see model_not_available_on_endpoint, the protocol is wrong, not the key.
On QCode
One QCode key works with all three protocols, and the request path decides which one is used: Claude goes only through the Anthropic endpoint https://api.qcode.cc/api; GPT models go through the OpenAI endpoint (Chat at https://api.qcode.cc/openai/v1) and cannot use the Anthropic endpoint; GLM, Kimi, DeepSeek and Qwen work on both. One key can call claude-sonnet-5-5, claude-opus-5-5, gpt-6.1-sol, glm-5.3, deepseek-v4.1-flash, kimi-k3, qwen3.8-max and more, billed per token, with each model's price on /models. Users in North America and Europe can switch the domain to us.qcode.cc (the Los Angeles node).
Frequently asked questions
Which AI coding tools support a custom API (BYOK)?
As of 2026-10-08, all 16 tools checked on this page do. GitHub Copilot, Cline, Roo Code, Kilo Code, Zed, Trae, JetBrains (Claude Code plugin), Claude Code, Claude Desktop, Cherry Studio, OpenCode, Aider, Crush and Droid can use the Anthropic protocol; for Cursor and Codex CLI, only an OpenAI-protocol custom address turned up in the official docs. The setting names for each tool are in the list above.
Can a tool that only supports OpenAI-compatible APIs use Claude?
Not on QCode. QCode serves Claude models only through the Anthropic endpoint; a claude- model sent to the OpenAI endpoint returns model_not_available_on_endpoint, and because that check runs before authentication, a different key does not help. For Claude, use a tool that supports the Anthropic protocol, or Claude Code; OpenAI-compatible tools still work normally with GPT and the Chinese model families.
Can Cursor use my own base URL?
Yes, for the OpenAI row: the official release notes give it a Base URL override switch, which works with GPT and the Chinese model families. As of 2026-10-08 no setting to change the Anthropic row's base URL was found in Cursor's official docs, so this list does not count Cursor as Claude-capable. The docs also say that with your own key every request passes through Cursor's servers for final prompt building, and Tab completion keeps using Cursor's built-in models. To use Claude inside Cursor, run Claude Code in the integrated terminal.
Can GitHub Copilot use my own key for Claude?
Yes. The VS Code Custom Endpoint supports the Anthropic Messages API, with the url set to the full address https://api.qcode.cc/api/v1/messages; for Copilot CLI set COPILOT_PROVIDER_TYPE=anthropic and COPILOT_PROVIDER_BASE_URL to https://api.qcode.cc/api. The VS Code docs state that BYOK powers chat, tools and MCP only, and inline suggestions still need a GitHub sign-in.
Should the base URL end in /api or /openai/v1?
It depends on the protocol: the Anthropic protocol takes https://api.qcode.cc/api, with no /v1 and no trailing slash; OpenAI-compatible takes https://api.qcode.cc/openai/v1. Exceptions follow each tool's setup page: GitHub Copilot in VS Code needs the full address https://api.qcode.cc/api/v1/messages, OpenCode uses https://api.qcode.cc/claude/v1, and the OpenAI type in Cherry Studio and Codex CLI take only https://api.qcode.cc/openai.
Can one key serve several tools at once?
Yes. The QCode docs say one API key works with all three protocols and the request path decides which; the Cursor setup page adds that Cursor's settings and Claude Code in the terminal authenticate separately but share quota when they use the same key. Every model is billed per token, with prices on /models.
Sources
Tool capabilities and setting names come from each tool's official docs, crawled 2026-10-08: the Claude Code environment variable, VS Code and JetBrains pages; the Claude Desktop third-party gateway page; the VS Code language model docs and the 1.122 release notes; the GitHub Docs page on Copilot CLI BYOK and the GitHub changelog; the Cursor help center BYOK page, the Agents Window release notes and the enterprise privacy page; the configuration docs of Cline, Roo Code, Kilo, Zed, Trae, Cherry Studio, OpenCode, Aider and LiteLLM, Crush, Factory and OpenAI Codex; and the Roo Code GitHub repository record. The QCode values come from the docs.qcode.cc tool compatibility overview, the endpoints and API formats page, the environment variables page and each tool's setup page, crawled the same day.
One key, plugged into the tools you already use
The Anthropic endpoint for Claude, the OpenAI-compatible endpoint for GPT and the Chinese model families, billed per token, with each model's price on /models.
Related reading
Cline with OpenAI-compatible APIs
BYOK setup in Cline, switching models, and how it compares with subscriptions.
Cherry Studio Claude API setup
Which provider type to pick, what address to enter, and common errors.
ANTHROPIC_BASE_URL setup guide
Pointing Claude Code at a custom endpoint with ANTHROPIC_BASE_URL and AUTH_TOKEN.
This page was checked on 2026-10-08; setting names and menu paths follow each tool's official docs and the docs.qcode.cc setup pages, tool updates may move these entry points, and upstream changes may happen without notice. Model availability is whatever /models shows.