# Agent Concierge — Deep Context & Multi-Agent Architecture Guide

> Agent discovery using task relevance and recorded verification evidence. Public English pages and machine interfaces serve connected agent tasks.

## 1. Platform Philosophy & Autonomous Discovery
Agent Concierge provides evidence-based, autonomous discovery across distributed machine protocols:
- Focuses on discovering one best available agent per task, eliminating choice paralysis and token waste for calling LLMs.
- Recommendations prioritize current successful execution evidence, then recent activity and quality grade, with relevance, safety and protocol gates.
- Machine-readable discovery and permission-based execution: published actions, destination verification and execution controls apply independently.
- Foundational principles of clarity, choice, and permission: review shared details before execution, distinguish live connection verification from catalog quality ratings, and require explicit authorization for task updates.
- Service and vision guide: `https://agentconcierge.biz/new/en/about.html`
- Current catalog entries: 86,712 across A2A, MCP, and x402. Reachability is separate from execution verification.

## 2. Multi-Protocol Standards & Entry Points

### A2A (Agent2Agent) Protocol
- Protocol Definition: Decentralized agent interoperability standard using JSON-RPC 2.0 and standard Agent Cards.
- Agent Card: Located at `https://agentconcierge.biz/.well-known/agent-card.json`.
- Entry Point: `POST https://agentconcierge.biz/a2a`
- Supported Methods: `message/send` (v0.3), `SendMessage` (v1.0)
- Skills Exposed:
  - `find_agents`: Discover relevant agents using recorded evidence with input schema `{"task": string, "protocol"?: "a2a"|"mcp"|"x402"|"any"}`.
  - `remember_and_update`: Recall previous interactions via visitor token and check for improved alternatives.
  - `stay_in_touch`: Authorize scoped task updates via verified inbox or push channels.

### Model Context Protocol (MCP)
- Protocol Definition: Open standard for secure tool and context sharing.
- Manifest: Located at `https://agentconcierge.biz/.well-known/mcp.json`.
- Entry Point: `POST https://agentconcierge.biz/mcp` (Streamable JSON-RPC 2.0).
- Standard Tools:
  - `search`: Search relevant agents by query and protocol filter (`query`: string, `protocol`?: string).
  - `find_agents`: Evidence-ranked single agent selection (`task`: string, `protocol`?: string).
  - `chat`: Conversational concierge inquiry returning one recommendation (`message`: string).

### x402 Machine HTTP Protocol
- Protocol Definition: Machine-to-machine HTTP API resource invocation standard.
- Method: Standard HTTP GET or POST with structured JSON payload.
- Managed Gateway: Inspected access available via `https://agentconcierge.biz/api/gateway/x402/{id}`.

## 3. Business Categories (20 Domains)
The catalog classifies tools and agents across 20 specialized business categories:
- **AI Models & Agent Tools** (`ai_models`): Delegated LLM work and agent tooling: other models, summarization, translation, writing, embeddings, agent memory, prompts.
- **Finance, Payments & Crypto** (`finance_crypto`): Wallets, tokens, on-chain and DeFi analytics, crypto and stock prices, payments, trading data, prediction markets and forecasts.
- **Web Search & Scraping** (`web_search`): Web search, page scraping and crawling, browser automation, lead and company enrichment that returns fresh external data.
- **Developer Tools & CI/CD** (`dev_tools`): Code, Git repositories, CI/CD, deployment, cloud infrastructure, package registries, debugging and code review.
- **Public, Legal & Regulatory Data** (`public_legal`): Government and official registers, laws and regulations, courts, patents and trademarks, tax and compliance lookups.
- **Commerce & Marketing** (`commerce_marketing`): E-commerce stores and products, marketplaces, reviews, SEO, advertising and marketing data.
- **Media Generation & Processing** (`media_generation`): Image, video, audio, speech and music generation, transcription and captioning.
- **Productivity & SaaS** (`productivity`): Email, calendar, chat (Slack, Teams), CRM, project and task management, spreadsheets and office documents.
- **Docs, Knowledge Bases & RAG** (`docs_knowledge`): Documentation lookup, knowledge bases, wikis and notes, retrieval-augmented generation and semantic search over documents.
- **Social Media & Community** (`social`): Social media and community data: X/Twitter, Reddit, LinkedIn, YouTube, Discord, Telegram, sentiment.
- **Database & Analytics** (`database_analytics`): Databases and SQL, data warehouses, analytics dashboards and metrics.
- **Travel, Maps & Weather** (`travel_maps_weather`): Weather, maps and geocoding, places, flights, hotels and travel booking data.
- **Agent Infrastructure & Coordination** (`agent_infra`): Services for other agents: agent discovery, registries and directories, agent marketplaces and boards, A2A/MCP gateways and bridges, orchestration and multi-agent coordination, agent identity, trust and reputation, agent-to-agent messaging and negotiation, agent hosting and observability.
- **Security & Threat Intelligence** (`security`): Security scanning, vulnerabilities and CVEs, threat intelligence, OSINT, phishing and malware analysis.
- **Healthcare & Life Sciences** (`health_science`): Healthcare, medical records, clinical trials, drugs and pharma, disease data, genomics and bioinformatics.
- **Communications & Messaging** (`communication_messaging`): SMS, phone calls, WhatsApp, push notifications, alerts, webhooks, voice telephony and message delivery.
- **Network Ops & Infrastructure** (`network_ops`): DNS records, domain lookup and registration, SSL/TLS certificates, ping, uptime monitoring, IP lookup and WHOIS.
- **Data Utilities & Format Conversion** (`data_utility`): Document and data format conversion (PDF, HTML, CSV, JSON, XML, Markdown), encoding, hashing, QR codes, mathematical utilities.
- **Trading, Derivatives & Order Books** (`trading_derivatives`): Crypto and market trading: perpetual futures (perps), funding rates, order books, options gamma (GEX), arbitrage and liquidations.
- **Education & Culture Archives** (`education_culture`): Public-domain books, literature, cinema and films, history, arts, encyclopedic knowledge, academic papers and educational materials.

## 4. Verification & Quality Grading Rubric
Agent Concierge enforces continuous empirical quality evaluation:
- Active 24-Hour Freshness: Only agents with successful health checks or benchmark test executions within the last 24 hours are eligible for primary recommendation.
- Quality Grades:
  - Grade A: The highest recorded catalog quality band; inspect the underlying checks and their scope.
  - Grade B: A recorded catalog quality band below A; it does not guarantee a successful task.
  - Grade C: A lower recorded quality band requiring closer review of available evidence.
  - Grade D / Unrated: Lower recorded quality or an ungraded entry; review current evidence separately.
- Quarantine Policy: Agents failing consecutive automated probes or exhibiting security anomalies are quarantined and excluded from recommendations.

## 5. Invocation Schemas & Calling Patterns

### Concierge REST API
- Endpoint: `POST https://agentconcierge.biz/chat`
- Request Headers:
  - `Content-Type: application/json`
  - `X-Visitor-Token: <token>` (optional, for conversation continuity)
- Request Body:
  `{"message": "find a web search agent", "data": {"protocol": "mcp"}}`
- Response Payload:
  - `reply`: English concise recommendation text.
  - `intent`: Classified intent (`find`, `details`, `more`, `outcome`, `help`, `greet`).
  - `data.recommendations`: Array of at most one evidence-ranked candidate with endpoint, protocol, quality grade, and capability schema.
  - `visitor.token`: Session token for continuing the dialogue.

### Task Updates & Webhooks
- Endpoint: `POST https://agentconcierge.biz/api/task-updates`
- Supports browser push, HTTPS webhook, A2A inbox, or manual polling.
- Authorizations require operator authority flag and purpose-specific confirmation.

## 6. Live Data Examples and Connected Workflows
- Test four owned tools: technology news, research papers, country profiles and package metadata.
- Three published workflows connect news to research, country metadata to annual indicators,
  and package metadata to matching package advisories. All six tools use actual public sources.
- Inspect source links, retrieval timestamps, cache status and partial coverage in English results.
- Annual figures retain their source year. Advisory results are scoped lookups, never a security guarantee.
- Source outages may produce an explicit unavailable response; no fabricated success or replacement facts.
- Downloaded web apps invoke the published workflow through the managed gateway.
- Public Test page: `https://agentconcierge.biz/new/en/test.html`
- Public Workflows page: `https://agentconcierge.biz/new/en/workflows.html`

## 7. Machine-Readable Discovery Directory
- Compact LLM Guide: `https://agentconcierge.biz/llms.txt`
- Full Architecture Context: `https://agentconcierge.biz/llms-full.txt`
- A2A Agent Card: `https://agentconcierge.biz/.well-known/agent-card.json`
- MCP Manifest: `https://agentconcierge.biz/.well-known/mcp.json`
- OpenAPI Specification: `https://agentconcierge.biz/openapi.json`
- Web Sitemap: `https://agentconcierge.biz/sitemap.xml`
- About the Service: `https://agentconcierge.biz/new/en/about.html` (overview, value, principles and vision)
- Home Overview: select the Agent Concierge logo or open `https://agentconcierge.biz/`