Skip to main content

​
OpenCLIs MCP Platform

MCP brings the data. OpenCLIs refines it for AI. OpenCLIs is a platform where developers register MCP services without deploying infrastructure. Authentication, hosting, caching, content refinement, and trust scoring — all provided. Developers bring their data; openclis makes it AI-ready.

​
The Problem

MCP responses are 70-90% metadata. AI agents pay tokens for UUIDs, timestamps, and formatting that carry no information. A Notion page costs 3,750 tokens through MCP — the actual content is 625 tokens.

​
The Solution

Any MCP Source → OpenCLIs Refiner → AI gets only what it needs

Notion (15,000 chars) → extract → index → cache → 200 tokens (95% saved)
GitHub (12,000 chars) → extract → index → cache → 150 tokens
Slack  (8,000 chars)  → extract → index → cache → 100 tokens

​
Platform Services

ServiceURLPurpose
Authauth.openclis.comOAuth 2.1 + developer registration
MCPmcp.openclis.comMCP protocol access (AI tools)
RESTrest.openclis.comREST API access (CLI, curl)
Registryapi.openclis.comCLI tools + domain trust (281 domains)

​
For Developers

Register a service in 4 API calls. No server deployment. No code changes to openclis. 
1. Register account     → API key
2. Register service     → MCP endpoint created
3. Define tools         → What your service does
4. Set backend URL      → Where requests go (your webhook)

Done. mcp.openclis.com/your-service is live.

​
Plugin System

Everything is data, not code:
  • Extractor rules: D1 templates that convert API responses to markdown
  • Handler types: kv_store, proxy, refiner, workspace, webhook
  • AI spec: Metadata that helps AI choose the right service
  • OpenAPI auto-generation: Register your API URL, tools are created automatically

​
Third-Party Auth

Your users authenticate with your auth system (Supabase, Firebase, Auth0, custom). openclis delegates validation to your URL. We never see your users’ credentials.

​
For AI Agents

Two ways to access any service: MCP (Claude Code, Cursor): 
POST mcp.openclis.com/workspace
→ JSON-RPC + schema load → ~1200 tokens
REST/CLI (Bash, curl): 
GET rest.openclis.com/workspace/search?query=MCP
→ plain JSON → ~200 tokens
Same data. Same auth. 6x fewer tokens via REST.

​
Trust Registry

281 domains scored for AI content trust. Ed25519 signed. 
doc.rust-lang.org    → 0.95 [verified,lang-docs]
stackoverflow.com    → 0.60 [community,qa]
medium.com           → 0.35 [external,blog]
AI agents check source trustworthiness before relying on external content.

​
Current Status

Live and operational. Not accepting public signups yet. What’s working:
  • 5 MCP services (niia, notion, consciousness, obsidian, workspace)
  • Notion + Obsidian unified workspace search
  • Supabase third-party auth integration tested
  • Webhook plugin system tested (weather service demo)
  • 281 trusted domains, Ed25519 verified
  • REST + MCP dual access

​
Early Access

If you’re building AI CLI tools or MCP services and want to test the platform, contact us: We’re looking for developers who want to:
  • Host MCP services without infrastructure
  • Reduce AI token costs with context refinement
  • Integrate existing APIs as MCP tools (OpenAPI auto-generation)
  • Test third-party auth integration