将 Cursor Web Docs 免费 API 转换为 OpenAI/Anthropic 兼容格式的代理服务。提供 Claude Code工具及图片支持。
1.8k
Stars
517
Forks
63
Open issues
6
Contributors
AI Analysis
Cursor2API is a proxy service that converts Cursor's free web documentation AI interface into OpenAI and Anthropic-compatible APIs, enabling integration with Claude Code, Cursor IDE, and third-party clients like ChatBox and LobeChat. It serves developers and organizations seeking to standardize AI model access across multiple tools without vendor lock-in. This project is specialized infrastructure for the Cursor IDE ecosystem rather than a general-purpose tool.
Inferred from signals mentioned in the README (tests, CI, type safety) — not a review of the actual code.
AI's overall editorial judgment — not an average of the bars above, can weigh other factors too.
Proxy that converts free Cursor AI docs API to OpenAI/Anthropic compatible format for Claude Code and Cursor IDE
Cursor2API is a TypeScript proxy that bridges Cursor's undocumented web API to standard OpenAI and Anthropic API formats. Built for developers using Claude Code or Cursor IDE who want to run these tools locally or integrate them into other clients. The project gained significant traction (~1,800 stars in 6 months) primarily within Chinese developer communities and power users seeking free access to Claude models through Cursor's infrastructure. It solves a real but fragile problem: Cursor's terms of service do not explicitly permit this usage pattern.
Created December 2025 during a period when Cursor offered free API access through its web documentation interface. The project emerged from the reverse-engineering community's discovery that Cursor's chat endpoint could be proxied. As of the README date (April 2026), Cursor deprecated the gemini-3-flash model, signaling API instability. This is a post-hoc wrapper around undocumented infrastructure, not a native implementation.
Rapid growth from zero to 1,834 stars in ~6 months suggests strong demand among cost-conscious developers and Claude enthusiasts in Asia. Growth has decelerated to 9 stars per week (as of June 2026), indicating either market saturation or increased platform friction. The project's Chinese documentation and cultural context suggest adoption concentrated in specific geographic/linguistic communities rather than organic cross-platform spread.
Adoption not verified through direct evidence (GitHub issues, public case studies, or company endorsements missing from README). Indirect signals: 1,834 stars and 518 forks suggest non-trivial install base, but stars alone do not confirm production usage. Chinese-language issues/discussions likely exist but not visible in English-only analysis. No reported deployment size, enterprise adoption, or SLA history provided. Project likely has quiet, distributed usage among individual developers and small teams, but scale is unquantifiable.
Based on README: appears to be an Express.js HTTP proxy with message format conversion (Anthropic ↔ OpenAI), tool parameter repair, vision downsampling (local OCR or API-based), and response sanitization. Implements streaming (SSE), compression strategies (3 configurable levels), token budget management, and a web-based log viewer. Likely uses a single-process Node.js architecture without clustering or distributed state; unclear if designed for horizontal scaling. README describes defensive engineering (50+ refusal patterns, multi-layer identity protection, truncation recovery) suggesting focus on robustness under adversarial conditions.
Not documented in README. No mention of test suite, CI/CD pipeline, or test coverage metrics. Production-readiness appears assumed rather than verified.
Last push 2026-06-01, ~2 weeks before evaluation date — actively maintained. README version indicates v2.7.8 with recent features (thinking support, context pressure inflation, adaptive history budgets) suggesting active development. Frequency of feature additions (multiple v2.x releases visible in version history) indicates engaged maintainer(s). However, only one known maintainer visible in repo metadata; single-person projects carry sustainability risk.
ADOPT IF: you are cost-conscious, comfortable with undocumented APIs, need Claude access in a non-commercial setting, and accept terms-of-service ambiguity. The technical implementation appears solid (streaming, compression, tool handling) and actively maintained. AVOID IF: you require legal certainty, production SLA, enterprise support, or guaranteed long-term API stability. Cursor2API depends entirely on Cursor's undocumented web endpoint—deprecations (like gemini-3-flash) suggest the foundation may shift. Also avoid if you use Cursor Pro; official integrations are preferable. MONITOR IF: you are evaluating this for small-scale team use; watch for Cursor policy changes, API breakage patterns, and whether Anthropic or Cursor issue cease-and-desist notices.
Independent dimensions
Mainstream potential
2/10
Technical importance
5/10
Adoption evidence
4/10
- Depends on undocumented, unsupported Cursor API endpoint. Cursor could revoke access at any time (precedent: gemini-3-flash deprecation in April 2026 already signals API churn).
- Likely violates Cursor's terms of service. No explicit permission from Cursor/Anthropic. Legal or account-ban risk unclear but non-zero.
- Single maintainer, no backup. Project sustainability uncertain if maintainer becomes unavailable.
- No published test suite or production deployment metrics. Reliability in high-volume or mission-critical scenarios unverified.
- Token budget and compression heuristics are emergent/ad-hoc (e.g., 'context pressure inflation' to trick client into compressing). Behavior may diverge from expected Claude API semantics under edge cases.
Cursor2API will likely remain a niche tool for cost-sensitive individual developers and small teams, particularly in Asia. Mainstream adoption is improbable because: (1) it depends on a closed API with no stability guarantees, (2) official Anthropic and OpenAI APIs are well-established and legal, and (3) the value proposition (free Claude access) disappears if Cursor monetizes or closes the endpoint. Most likely trajectory: slow decline as Cursor tightens security and developers migrate to paid official APIs. Possible brief resurgence if Cursor remains generous, but unlikely to reach broader industry adoption.
Newsletter
Get analyses like this every Monday
Free weekly digest of the most interesting open-source discoveries.
Languages
Information
- Website
- https://cursor.com/cn/docs
- Language
- TypeScript
- License
- MIT
- Last updated
- 1mo ago
- Created
- 7mo ago
- Analyzed with
- anthropic/claude-haiku-4-5
Stars over time
Contributors over time
Top 100 contributors only — repos with more will plateau at 100.
Open issues
Open pull requests
No open pull requests.
Top contributors
Similar repos
cursor/cursor
Cursor is an AI-powered code editor that serves developers seeking IDE features...
oslook/cursor-ai-downloads
This repository maintains a curated collection of official Cursor AI download...
ding113/claude-code-hub
Claude Code Hub is an API proxy and gateway service built with Next.js 15,...
| Repository | Stars | Week Δ | Language | Score | Updated |
|---|---|---|---|---|---|
|
7836246/cursor2api
this
|
1.8k | +13 | TypeScript | 7/10 | 1mo ago |
|
cursor/cursor
|
33k | — | — | 6/10 | 2mo ago |
|
oslook/cursor-ai-downloads
|
3.2k | — | TypeScript | 7/10 | 2d ago |
|
ding113/claude-code-hub
|
3.2k | — | TypeScript | 7/10 | 4d ago |
 siteboon/claudecodeui
|
12.5k | — | TypeScript | 7/10 | 16h ago |
 decolua/9router
|
21.6k | — | JavaScript | 7/10 | 1h ago |
Cursor is the source being proxied. Cursor2API is a workaround, not a replacement. Cursor's official API remains closed; this project exploits web endpoint leakage.
Similar spirit (Claude access wrapper), but appears to focus on UI rather than API proxying. Cursor2API is more infrastructure-oriented.
Likely another Cursor/Claude wrapper. Direct comparison not possible without inspection, but ecosystem fragmentation suggests multiple reverse-engineer projects competing for same niche.
Legitimate alternative: use official Anthropic API with paid credits. Cursor2API is cheaper (free tier dependent), less stable, and likely violates Cursor ToS.
Alternative strategy: run open-source models locally. Cursor2API provides Claude-level quality at risk; LocalAI offers legality and permanence at lower capability.
