find-agent-service（寻找 Agent 服务）

Skill: find-agent-service

Use this skill whenever a user or agent needs to identify the right agent-native service for a particular task. Beyond finding a service, always surface how the agent actually starts using it — because in some cases, the answer is a single sentence the agent can execute right now.

The most important concept: URL Onboarding

Some services in this catalog can be joined by an agent with a single instruction, right now, with no human setup:

Read <url> and follow the instructions.

This is called URL Onboarding — the service hosts a machine-readable skill/protocol file that an agent reads and follows to self-register. The agent becomes part of the service's ecosystem autonomously. This is qualitatively different from SDK integration (which requires coding) or MCP (which requires config file changes).

Services with URL Onboarding (highest priority to surface):

When a task maps to one of these services, always lead with the onboarding instruction — it's the most actionable thing you can give an agent.

When to activate

Activate this skill when the user asks things like:

• "What service should my agent use for email?"
• "Is there an agent-native payment API?"
• "How can my agent browse the web?"
• "I need my agent to remember things across sessions — what do I use?"
• "What's the best way for an agent to approve a high-risk action?"
• "How does my agent join Moltbook / Ensue / autoresearch?"
• "What can my agent do right now, with no setup?"

Category map

How to find the right service

Step 1 — Map the task to a category

Use the table above. Note the onboarding pattern — if it's URL Onboarding, you can give the agent a one-sentence instruction immediately.

Step 2 — Read the category file

The catalog is at services/{category}/README.md. Read it to see all services and their onboarding commands.

Category folder names (15 categories):

• services/communication/
• services/browser-and-web-execution/
• services/tool-access-and-integration/
• services/oversight-and-approval/
• services/commerce-and-payments/
• services/agent-runtime-and-infrastructure/
• services/memory-and-state/
• services/search-and-web-intelligence/
• services/code-execution/
• services/observability-and-tracing/
• services/durable-execution-and-scheduling/
• services/meeting-and-conversation/
• services/voice-and-phone/
• services/llm-gateway-and-routing/
• services/agent-social-network/

Step 3 — Read the service file

Each service has a detailed file at services/{category}/{service-name}.md containing:

• How to Use (Agent Onboarding) — the quickest entry point (always check this first)
• Primary primitives (the agent-specific abstractions)
• Protocol surface (SDK, REST API, MCP, webhooks)
• Agent Skills install command
• MCP server details
• Use cases with concrete examples

Step 4 — Recommend with the right entry point

Match the recommendation to the onboarding pattern:

## Recommended service: {Service Name}

**Why:** {specific primitive that matches the task}

**How to start** ({URL Onboarding / Coding-time Skill / MCP / SDK / Daemon}):
{one-line instruction appropriate to the pattern}

URL Onboarding example:
  Read https://www.moltbook.com/skill.md and follow the instructions to register and join.

Coding-time Skill example:
  npx skills add tavily-ai/skills

MCP example:
  Add to mcp_servers: { "command": "npx", "args": ["-y", "bb-browser", "--mcp"] }

SDK example:
  pip install mem0ai  # then: m.add(messages, user_id="agent-1")

**Relevant use case from the catalog:**
> {quote the use case that matches the task}

When nothing fits

If no service in the current catalog fits the task:

1. Say so clearly — do not recommend an agent-adapted service as if it were agent-native.
2. Note the closest existing service and explain what is missing.
3. Suggest opening a new service issue if the user knows of a qualifying service.

Classification reminder

This catalog only lists agent-native services. Do not recommend:

• agent-adapted services (e.g., Resend, Stripe, Twilio) — built for humans, agent layers added later.
• agent-builder platforms (e.g., Dify, n8n, LangGraph) — for humans building agents.

If asked about those, explain the distinction and point to the Excluded / Boundary Cases section in README.md.