SocQ Social and SEO Research

Research social and SEO data

Use SocQ to collect public social and SEO data through an asynchronous, credit-metered workflow.

Choose the integration

1. In OpenClaw, prefer the socq CLI when it is installed. Otherwise run the same commands through npx @socq/cli.
2. Prefer the hosted MCP server at https://api.socq.ai/mcp when the client already has SocQ MCP configured.
3. Use a filtered MCP URL when the platforms or endpoints are known:
   • ?platforms=youtube,tiktok for at most five platforms.
   • ?tools=youtube_comments,x_search for at most thirty endpoint tools.
4. Run npx @socq/mcp when the client only supports local stdio MCP.
5. Use REST only when neither CLI nor MCP is available.
6. Attribute every execution started by this Skill:
   • pass _request_source: "skill" to compact or typed MCP execution tools;
   • pass --request-source skill to CLI execution commands;
   • send X-Socq-Source: skill-rest for direct REST execution.

Never place an API key in a prompt, query string, committed file, or shell command that will be retained. Read authentication.md before configuring credentials.

SocQ is an external, credit-metered service. A SocQ account and SOCQ_API_KEY are required, and research requests may consume paid credits. Check the available credits and expected endpoint cost before starting a large or cross-platform request.

Execute research

1. Restate the requested platforms, entities, date range, and result volume.
2. Search the Capability Registry, then describe the selected endpoint before constructing input. Read catalog.md when tool discovery is unavailable.
3. Read the matching platform reference before choosing inputs or constructing a payload: Facebook, Facebook Ad Library, Facebook Marketplace, Instagram, LinkedIn, Pinterest, Reddit, Threads, TikTok, TikTok Shop, X, or YouTube. For keyword research, intent, organic results, or site rankings, read SEO.
4. Prefer direct URLs or canonical usernames over broad social search when the user supplies them. For SEO work, preserve the requested market and language; do not compare keyword metrics from different locations as if they were equivalent.
5. Check account credits before a large or cross-platform run. Read billing.md for cost controls.
6. Submit with a reusable idempotency key when a transport retry is possible.
7. Treat queued and running as incomplete. Poll until succeeded or failed; follow async-tasks.md.
8. Follow every next_cursor needed for the requested scope, respecting a user-provided result cap. Read pagination.md.
9. Use task files for complete raw JSONL exports when normalized pages are insufficient.
10. Report partial coverage, failed platforms, filters, and collection time with the results.

For research spanning multiple networks, follow cross-platform.md. For authentication, rate limits, provider failures, and recovery, follow errors.md.

Guardrails

• Collect only public data supported by the selected endpoint.
• Do not retry a failed paid request blindly; inspect the normalized error first.
• Do not invent unsupported parameters. Re-read the endpoint schema after a validation error.
• Do not claim completeness when pagination stopped early, a provider failed, or the requested date filter is unsupported.
• Keep task IDs in the working notes so interrupted research can resume without resubmission.