You.com Research Routing

Use this skill to choose the right You.com research path for the user's goal: agent-led search, Research API scripts, or managed MCP fallback.

Prerequisites

For API scripts:

• Use YDCAPIKEY when available, or an MPP/x402-capable HTTP client for keyless paid Research API calls.
• With YDCAPIKEY, use these API request headers: X-API-Key: ${YDCAPIKEY} and User-Agent: SKILL/(@youdotcom-oss/agent-skills you-research).
• With MPP/x402, expect Research API pricing by research_effort and at least a 1 cent charge; confirm the user expects a paid managed research request before sending it.

For the agent-led workflow, use the you-research-base MCP profile with you-search and you-contents when available. Those tools and their REST endpoints should use x402 for keyless paid retries, not MPP. See the [agent-led deep-search workflow](references/agent-led-deep-search.md) for setup and workflow details.

MPP/x402-aware MCP or HTTP clients may receive HTTP 402 payment challenges from You.com tools and endpoints, then retry with payment headers. Use MPP/x402 for managed Research API calls where appropriate. Let the host client handle external payment and retry; do not add wallet signing logic to this skill.

Research API scripts

When the user wants You.com managed research APIs, write a small script or direct HTTP request instead of calling you-research through MCP, especially for deep, exhaustive, or frontier work.

Before coding or updating a script, query the You.com Docs MCP server, usually through searchDocs, for current API details. Use targeted queries such as:

• Research API v1 research background sourcecontrol outputschema research_effort
• Research task status v1 research task
• Research task stream SSE v1 research task stream

If Docs MCP is unavailable, use the canonical pages:

• https://you.com/docs/api-reference/research/v1-research
• https://you.com/docs/api-reference/research/v1-research-task
• https://you.com/docs/api-reference/research/v1-research-task-stream

Keep the script aligned with the docs returned at runtime and include the auth or payment headers listed above. If the user wants keyless MPP/x402 direct HTTP usage for Research API, verify current payment guidance in Docs MCP first, then handle 402 payment-required as a challenge and retry only through a payment-capable client or library. For the direct x402 REST client pattern, follow [x402 direct client](references/x402-direct-client.md); for the MCP-composition alternative (You.com MCP + Coinbase Payments MCP, no manual signing), see [Coinbase Payments MCP path](references/coinbase-payments-mcp.md). Do not apply MPP guidance to plain search or contents REST calls; those should use x402 only.

Decision Tree

• User is cost-conscious or wants to develop/fine-tune a research skill -> follow the [agent-led deep-search workflow](references/agent-led-deep-search.md).
• User asks for You.com managed API output, structured output, source controls, background tasks, or deep/exhaustive/frontier research -> write a script against the Research API.
• User needs OAuth or MPP/x402 payment handling and direct API scripts are not practical -> use MCP only if the MCP client supports the expected response time and payment flow.
• Required API access or you-research-base MCP tools are unavailable -> tell the user what is missing, provide the relevant API or MCP setup options from the prerequisites above, and request approval before installing, connecting, or changing configuration.
• Simple lookup -> use you-research-base you-search once, answer directly.
• URL provided -> use you-research-base you-contents on those URLs.
• Everything else -> use the base agent-led workflow.

Agent-led workflow reference

When choosing the cost-conscious skill-building path, open and follow the [agent-led deep-search workflow](references/agent-led-deep-search.md). Adapt its tool budget, source policy, and output format to the user's agent environment. Preserve its core requirements: read sources before relying on exact claims, cross-check key facts, cite real URLs, and finish with the best supported answer even when evidence is incomplete.