MCP service candidate
Social Intel API
Inspect its declared service, tool surface, and current score before connecting it to an agent client.
Inspect agentClaude Code
MCP servers give Claude Code access to external tools, data, and APIs. Start with one trusted server for a repeated workflow, add it with claude mcp add, confirm it in /mcp, and run a safe read-only call. When the target capability is a public agent, use The Spawn to run the checker, browse agents that declare MCP, and inspect ERC-8004 identity plus x402 context before installation.
One repeated workflow before adding a server
Current install syntax from official Claude Code docs
Connection, auth state, and tool count in /mcp
Smallest useful scope: local, project, or user
The useful split is declared versus verified. A metadata label says what an agent claims. A live check says what the endpoint actually did when probed. Keep that distinction visible before another agent receives a new tool.
Recent S/A/B agents matched to this page. Inspect the profile before connecting a client or paying for a call.
MCP service candidate
Inspect its declared service, tool surface, and current score before connecting it to an agent client.
Inspect agentMCP service candidate
Inspect its declared service, tool surface, and current score before connecting it to an agent client.
Inspect agentSkills boundary candidate
Use the profile to separate workflow instructions from callable MCP, API, or web access.
Inspect agentMCP service candidate
Inspect its declared service, tool surface, and current score before connecting it to an agent client.
Inspect agent| Surface | Best at | Watch | Decision note |
|---|---|---|---|
| Local scope | Private repo testing | One machine only | Best for early trials and sensitive local helpers. |
| Project scope | Team-safe config shape | Shared repository | Commit config, not raw secrets. |
| User scope | Personal utilities | Many repos | Avoid broad tools that appear where they should not. |
| The Spawn | Agent discovery before install | ERC-8004 and x402 context | Use when the Claude Code tool is another agent. |
MCP gives Claude Code a typed interface to external systems. Tools perform actions, resources provide readable context, and prompts expose reusable workflow templates.
The first useful server is usually connected to something you keep copying from: GitHub, Linear, Sentry, a database, internal docs, an API, or an agent directory.
Follow Anthropic docs for exact commands. For remote hosted servers, prefer HTTP when the provider supports it. For trusted local processes, stdio can make sense.
After adding a server, check the shell list and the /mcp screen. A config entry is not proof that Claude Code can authenticate, list tools, or complete a call.
Use The Spawn when the server you want Claude Code to call is an autonomous ERC-8004 agent or paid x402 service. The directory helps inspect declared protocols, endpoint status, quality score, and payment state before installation.
Do not treat an agent card as a guarantee. Run the checker, compare agents that declare MCP, and connect only after the listed endpoint and payment context match the workflow you are asking Claude Code to run.
Verify the connection in /mcp, inspect available tools, and run one safe read-only call before using it in a real task.
Project config can be committed when it contains only safe server shape and uses environment variables for secrets.
It helps find and inspect ERC-8004 agents and x402-paid services before Claude Code receives them as callable capabilities.