Welcome to MercurAI Workspace

Leverage our dual-engine architecture: The Data Engine for grounded semantic search, and the WebMCP Studio for autonomous agent actuation.

WebMCP Studio (Actuation)

Transform your frontend into a W3C-compliant agent execution layer, enabling Chrome Copilots to submit forms and trigger actions.

Data Engine (AI Search)

Feed global AI Search platforms and RAG pipelines with clean, vectorized semantic data via pgvector.

Authentication & Protocol Access

Your access is governed strictly by the x-api-key header linked to your active subscription tier.

API Gateway

Route all standard HTTP requests through the Uvicorn-powered tollbooth. Standard REST endpoints are available here.

Base URL: https://dev-mcp.mercurai.online/v1/

Port: 8080

FastMCP Stream (SSE)

Establish persistent connections for autonomous agents via Server-Sent Events (SSE) to bypass traditional REST limitations.

curl -N -H "x-api-key: YOUR_KEY" \
  https://dev-mcp.mercurai.online/v1/sse

WebMCP Studio (Agent Actuation)

Aligned with the W3C Web Machine Learning Community Group draft specification, this protocol enables autonomous agents to safely trigger deterministic functions on your site (bypassing brittle visual DOM scraping). This is explicitly for Agent Actuation, not passive AI Search indexing.

1

Domain Registration

Register your target URL in the WebMCP Portal and verify ownership.

2

Environment Profiling

The mercurai-scout.js script evaluates site complexity, detecting frameworks (React/Vue) and Shadow DOMs to determine agent execution readiness.

3

W3C Draft Schema Generation

Use the mercurai-discovery.js probe in your console to scan forms and output JSON Schema (Draft-07) compliant definitions required for W3C tool registration.

4

Browser Native Registration

The runtime bridge utilizes the proposed document.modelContext.registerTool() API to expose capabilities natively, triggering built-in browser permission dialogs for maximum security. A polyfill is provided for legacy clients.

Data Engine & Vector Retrieval (AI Search)

This is the indexing and discovery layer. Access raw, structured market intelligence driven by FastEmbed (BAAI/bge-small-en-v1.5) and pgvector for AI Search engines and RAG pipelines.

Semantic Lookup

Execute sub-millisecond similarity queries directly via the /v1/dataset/export endpoint.

Bulk Export

Filter datasets by industry niche, city, or service, and export up to 5,000 records per request as JSON or CSV.

Market Ingestion

The ingestion engine cleans HTML/PDFs, generates fresh embeddings, and inserts them into pgvector.

$ python run_market_ingestion.py --source ./data

AI Agent Configuration

Connect external AI models to your authenticated tools to execute live semantic searches and actuation protocols. Note: Native WebMCP agents (like Chrome 149+) discover tools automatically without this config file.

Claude Desktop Integration (Fallback Mode)

For external automation harnesses relying on the traditional MCP protocol rather than the W3C WebMCP bridge, add the SSE URL and your x-api-key header to your claude_desktop_config.json.

{
  "mcpServers": {
    "mercurai-workspace": {
      "command": "curl",
      "args": [
        "-N",
        "-H",
        "x-api-key: YOUR_API_KEY",
        "https://dev-mcp.mercurai.online/v1/sse"
      ]
    }
  }
}

Universal Pipeline Test

Run the CLI script to verify your agent handshake and test tool execution against the database.

$ python run_agent_test.py --agent claude --test semantic_search