Havoptic API Documentation
All Havoptic data is publicly accessible without authentication. Use these endpoints to programmatically access release data, feature comparisons, and feeds for AI coding tools.
Developer docs with a quickstart, API keys and rate limits, pagination, typed changes, webhooks, MCP setup and the error catalog: https://havoptic-prod-web.pages.dev/docs.
Why Havoptic's API
Built for depth on AI coding tools rather than breadth across every changelog.
- Typed change items with evidence - Each release is split into typed changes (feature, fix, breaking, deprecation, security, and more), and every item carries a verbatim quote from the source notes plus a confidence score. Typed changes cover releases from 2026-10-19 onward; older releases report changes_available: false.
- Cross-tool feature intelligence - Who shipped a feature first, how long others took to follow, and the mechanics underneath, compared across every tracked AI coding tool.
- Upgrade reports - Breaking changes, deprecations, and migration notes rolled up for any version range via /api/v1/tools/{id}/upgrade or the get_upgrade_notes MCP tool.
- Agent-first ergonomics - MCP tools with token budgets, markdown output, citation links, and freshness metadata, plus OpenAPI 3.1 and JSON Schemas for typed clients.
- Provenance - Raw release notes are retained and hashed, and every AI-derived field is labeled with the model and prompt version that produced it.
MCP Server
Havoptic is queryable directly from AI assistants via the Model Context Protocol. Remote MCP server: https://mcp.havoptic.com/mcp (streamable HTTP transport, read-only, no authentication). Connect from Claude Code: claude mcp add --transport http havoptic https://mcp.havoptic.com/mcp
MCP tools: list_tools, get_latest_releases, whats_changed_since, search_changes, get_release, get_upgrade_notes, get_breaking_changes, compare_tools, feature_adoption, cadence_stats; deprecated aliases (kept until at least 2027-11-01): list_tracked_tools → list_tools, latest_releases → get_latest_releases, release_history → get_latest_releases with since, search_release_notes → search_changes, feature_timeline → feature_adoption.
Official MCP Registry name: com.havoptic/releases (https://registry.modelcontextprotocol.io/v0/servers?search=com.havoptic/releases). Server card: https://www.havoptic.com/.well-known/mcp.json.
MCP resources: havoptic://tools, havoptic://tools/{id}, havoptic://releases/{id}, havoptic://tools/{id}/changelog?since=, havoptic://schema/{name}, havoptic://docs/api. MCP prompts: weekly_ai_tools_brief, upgrade_check, compare_for_team, release_announcement.
REST API (v1)
- GET /api/v1/releases - Filterable, paginated release list. Params: tool, type, since, until, q, limit (max 100), offset. Cache: 5 min.
- GET /api/v1/releases/{id} - One release by id with full release notes.
- GET /api/v1/tools - Tracked tools with release counts and latest release.
- GET /api/v1/tools/{id}/cadence - Release cadence stats (7/30/90-day counts, average gap).
- GET /api/v1/compare?tools=a,b - Cadence, feature differences, and editorial verdict for a tool pair.
- GET /api/v1/feature-timeline - Who shipped each notable cross-tool feature first, with follower lag.
- GET /api/v1/mechanics - Behaviors and the machinery underneath named features, joined with first-ship dates.
Bulk Datasets and Feeds
- GET /data/releases.json - All tracked AI coding tool releases with version, date, summary, and full notes. Updated daily. Cache: 1 hour.
- GET /data/feature-matrix.json - Feature comparison matrix across all tracked tools. Cache: 24 hours.
- GET /data/mechanics.json - Human-curated behaviors and mechanics taxonomy.
- GET /feed.xml - RSS 2.0 feed of the 50 most recent releases. Cache: 1 hour.
- GET /feed.json - JSON Feed 1.1 of the 50 most recent releases. Cache: 1 hour.
- GET /llms.txt - Structured site description for AI assistants. Cache: 24 hours.
- GET /llms-full.txt - Extended LLM context with feature matrix, recent releases, and FAQ. Cache: 1 hour.
- GET /.well-known/agents.json - Machine-readable agent discovery file. Cache: 24 hours.
- GET /.well-known/mcp.json - MCP server card (endpoint, transport, tools, resources, prompts). SEP-2127 AI Catalog: /.well-known/ai-catalog.json. Cache: 1 hour.
Tool Identifiers
Use these IDs to filter releases by tool: claude-code, openai-codex, cursor, gemini-cli, kiro, github-copilot, windsurf, antigravity-cli, grok-build, cline, opencode, zed, factory-droid, devin, lovable, amp, copilot-vscode, replit-agent, bolt, v0, openai-platform, anthropic-api, gemini-api, xai-api, mistral-api, deepseek-api, amazon-bedrock, gemini-agent-platform, mcp-spec, openai-agents-python, openai-agents-js, claude-agent-sdk-python, claude-agent-sdk-typescript, vercel-ai-sdk, langgraph-python, langgraph-js, llamaindex, crewai, mastra, cloudflare-agents-sdk, ollama, vllm, lm-studio, huggingface-transformers.
Rate Limits
60 requests per minute. No authentication required. Please cache responses where possible.
For AI Agents
Live queries: MCP server at https://mcp.havoptic.com/mcp. Machine-readable discovery files are available at /.well-known/agents.json, /llms.txt, and /llms-full.txt. OpenAPI spec: /openapi.json.