A remote MCP server that ships skills (procedural knowledge) — not tools — to any connected AI agent. Hosted on Cloudflare Workers.
The premise: filesystem SKILL.md files (the agentskills.io standard) are great, but they're static, install-per-harness, can't auth-gate per user, and can't carry live data. SkillStream is the service layer above filesystem skills.
V0 is a public MCP endpoint that any compatible client connects to and immediately gets a battle-tested methodology skill set: brainstorm-first → plan-before-action → validate-before-execute → verify-after-complete, plus self-healing-recovery and token-conscious-iteration.
GET /— JSON info (skill list, copy-paste client configs)POST /mcp— MCP Streamable HTTP (current spec)GET /sse— MCP SSE (legacy)
Add this URL as a remote MCP connector:
https://skillstream-mcp.prbotsai.workers.dev/mcp
Or point your agent at the info endpoint for copy-paste configs:
https://skillstream-mcp.prbotsai.workers.dev/
9 skills exposed as MCP resources + prompts + tools:
skill://using-skillstream— start hereskill://brainstorm-firstskill://plan-before-actionskill://validate-before-executeskill://execute-with-disciplineskill://verify-after-completeskill://self-healing-recoveryskill://token-conscious-iterationskill://find-the-right-skill
Descriptions teach the agent when to load each one. Bodies load on demand. Total context cost at idle: ~1.5 KB.
- Drop
src/skills/your-skill-name.md - Add an import + entry in
src/registry.ts - Push to main — Cloudflare auto-deploys
Agent ──MCP/HTTP──► Cloudflare Worker (Durable Object per session)
├─ Resources: skill://<name>
├─ Prompts: skill_<underscored_name>
└─ Tools: search_skills, list_all_skills, activate_skill
V0 ships skills statically. V1 adds D1 + R2 + Vectorize + a hydration engine that injects live data into skill bodies at request time.
MIT.