Model Context Protocol
Ask For Agent Decisions, Not Raw Records
The MCP server is the task-first layer on top of the indexed registry. Use it to vet an agent before integrating, shortlist candidates by service, or resolve concrete endpoints without scraping the UI or stitching raw records together yourself.
Why MCP Instead Of REST?
Different surface, different job.
Use the REST API when you already know which record or endpoint you need. Use MCP when you want the model to help evaluate, compare, and decide.
REST API
Raw records
Fetch agents, stats, and feedback when your app wants structured data and you already control the decision logic.
MCP
Prompt-driven decisions
Ask for a recommendation, shortlist candidates, or resolve an integration path from a single task-oriented tool call.
Live MCP Workflows
Ask for a decision, not a raw record.
Start with one task. Run the live tool call, inspect the answer preview, and only drop into raw JSON if you actually need the transport details.
Vet Agent
Ask the MCP server for a recommendation, not just a record dump. It will summarize readiness, risks, service fit, and provenance.
Target Service
Question
Vet agent #28513 for MCP integration. Tell me if it looks ready, why, and what I should watch out for.
Advanced: raw tool request
{
"jsonrpc": "2.0",
"id": "demo-vet-agent",
"method": "tools/call",
"params": {
"name": "vet_agent",
"arguments": {
"agentId": 28513,
"targetService": "mcp"
}
}
}Live Result
Run the workflow to fetch a real MCP response.
Answer Preview
Run the workflow to see a live MCP response.
Advanced: raw MCP response
No response yet.Connection
Endpoint
Transport
Streamable HTTP
Auth
Optional Bearer token
Operating Mode
Stateless Streamable HTTP, typed inputs, and responses backed by the same indexed registry model that powers the explorer. Lead with recommendations and workflows, then drop into raw transport only when you need it.
Client Config
Claude Desktop
{
"mcpServers": {
"8004": {
"command": "npx",
"args": [
"mcp-remote",
"https://8004.fyi/mcp"
]
}
}
}Client Config
Cursor
{
"mcpServers": {
"8004": {
"url": "https://8004.fyi/mcp"
}
}
}Tool Reference
Five focused tools.
The MCP layer stays narrow on purpose: vet, browse, inspect, measure, and resolve.
Tool
vet_agent
Assess whether a specific agent looks ready to integrate, with a verdict, reasons, risks, service match, and provenance.
Key Params
agentId, targetService
Tool
search_agents
Search and browse agents by keyword, service, reputation, page, and sort.
Key Params
query, service, hasReputation, sort, page, limit
Tool
get_agent
Return the full agent record with metadata and optional reputation details.
Key Params
agentId, includeReputation
Tool
get_stats
Return registry-wide counts and sync metadata.
Key Params
none
Tool
resolve_endpoint
Find agents exposing a given service type and return matching endpoints.
Key Params
service, query, limit