# Chiark

> Cross-protocol quality index for AI agent services. Discovers, probes, and scores 2,000+ agents across A2A and MCP ecosystems from 9 registries.

Chiark measures the operational reliability of deployed AI agent endpoints — availability, protocol conformance, and performance — through automated three-tier probing every 30 minutes. It is the independent quality layer for agent-to-agent routing.

Live at https://chiark.ai

## API

The REST API is at `https://chiark.ai/api/v1`. All responses are JSON. No authentication required.

- [API Documentation](https://chiark.ai/docs): Interactive OpenAPI/Swagger docs
- [OpenAPI Spec](https://chiark.ai/openapi.json): Machine-readable API schema
- [Full API Reference](https://chiark.ai/llms-full.txt): Complete endpoint documentation

### Discover Agents (recommended for routing)

Find agents matching quality constraints, optionally filtered by task description:

```
GET /api/v1/discover?task=web+scraping&min_uptime=0.9&protocol=mcp&page_size=3
GET /api/v1/discover?min_score=80&protocol=mcp           (constraint-only, no task required)
GET /api/v1/discover?payment_enabled=true                 (all payment-enabled agents)
```

Response:
```json
{
  "agents": [
    {
      "id": "uuid",
      "name": "Agent Name",
      "operational_score": 87,
      "max_score": 100,
      "uptime_30d": 0.99,
      "p95_latency_ms": 245,
      "conformance": "pass",
      "protocol": "mcp",
      "endpoint_url": "https://...",
      "categories": ["Developer Tools"]
    }
  ],
  "total": 42
}
```

### Check Agent Status

```
GET /api/v1/agents/{id}/status
```

Response:
```json
{
  "agent_id": "uuid",
  "status": "operational",
  "is_alive": true,
  "http_status": 200,
  "response_time_ms": 145.2,
  "tls_valid": true,
  "probe_tier": 3,
  "timestamp": "2026-03-26T12:00:00Z"
}
```

Status values: `operational`, `degraded`, `reachable`, `offline`, `auth_required`, `unknown`

### Get Agent Capabilities

```
GET /api/v1/agents/{id}/capabilities
```

Response:
```json
{
  "agent_id": "uuid",
  "name": "Agent Name",
  "protocol": "mcp",
  "auth_required": false,
  "skills": [
    {"name": "translate", "description": "Translate text between languages", "tags": ["translation"]}
  ],
  "endpoint_url": "https://..."
}
```

### Get Response Sample

See what an agent actually returns — tool schemas, parameters, and pricing:

```
GET /api/v1/agents/{id}/sample
```

Response:
```json
{
  "agent_id": "uuid",
  "has_sample": true,
  "probe_tier": 3,
  "tool_count": 14,
  "tools": [
    {
      "name": "translate_text",
      "description": "Translate text between languages",
      "parameters": {"type": "object", "properties": {"text": {"type": "string"}, "target_lang": {"type": "string"}}},
      "annotations": {"readOnlyHint": true}
    }
  ],
  "server_info": {"name": "agent-server", "version": "1.0"},
  "x402_pricing": null,
  "last_probed": "2026-03-28T19:22:11Z"
}
```

### Report Routing Outcome

```
POST /api/v1/feedback
Content-Type: application/json

{"agent_id": "uuid", "success": true, "task_category": "translation"}
```

### Other Endpoints

- `GET /api/v1/rankings` — leaderboard with constraint filters (min_score, min_uptime, max_latency_ms, auth_required, payment_enabled, protocol, category)
- `GET /api/v1/agents/search?q=keyword` — search agents by name, description, skills
- `GET /api/v1/agents/{id}` — full agent detail with score breakdown and probe history
- `GET /api/v1/agents/{id}/sample` — probe response sample (tools, parameters, schemas, pricing)
- `GET /api/v1/agents/{id}/feedback` — aggregated routing success rate for an agent
- `POST /api/v1/register` — self-register an agent for indexing (endpoint_url or agent_card_url)
- `GET /api/v1/categories` — all categories with agent counts
- `GET /api/v1/stats` — ecosystem totals and averages

### Constraint Filters

- `min_score` (0-100): Minimum operational score
- `min_uptime` (0-1): Minimum 30-day uptime (e.g., 0.99 = 99%)
- `max_latency_ms`: Maximum P95 latency in milliseconds
- `auth_required` (true/false): Filter by authentication requirement
- `payment_enabled` (true/false): Filter by x402 payment support
- `protocol` (a2a/mcp): Filter by protocol
- `category`: Filter by skill category

## MCP Server

Chiark is available as an MCP server for agent discovery from Claude, Cursor, or any MCP client.

**Hosted endpoint (no install):** `https://chiark.ai/mcp/` — connect directly from any MCP client via HTTP transport.

**Standalone package:** `pip install chiark-mcp` ([PyPI](https://pypi.org/project/chiark-mcp/) / [GitHub](https://github.com/chiark-ai-dev/chiark-mcp))

Listed on [Smithery](https://smithery.ai) and [a2aregistry.org](https://a2aregistry.org).

Tools: `find_agent` (with constraints), `check_agent_status`, `get_agent_score`, `report_outcome`, `get_ecosystem_stats`

## Scoring

Operational Score = Availability (0-30) + Conformance (0-30) + Performance (0-40)

- Availability: 30-day uptime percentage
- Conformance: Agent Card valid + protocol response valid + endpoint match
- Performance: P95 latency + task completion rate + consistency bonus
- Auth-gated agents max score: 45/100 (can't verify performance without access)

## Data Sources

9 crawlers run every 24 hours: a2aregistry.org, MCP Registry, Smithery, Solana ERC-8004, awesome-a2a, GitHub topics, well-known endpoints, PulseMCP (when API key configured), alternative registries.

## Popular Categories (by agent count)

- AI & Agents (365+)
- Developer Tools (269+)
- Crypto & Web3 (188+)
- Cloud & Infrastructure (173+)
- Data & Analytics (118+)
- Productivity (113+)

## Optional

- [Methodology](https://chiark.ai/methodology): Full scoring methodology
- [Stats](https://chiark.ai/stats): Ecosystem analytics
- [Agent Card](https://chiark.ai/.well-known/agent.json): A2A Agent Card for Chiark itself