Find your landing page message
Connect a product’s value to a specific audience and write a clear first-screen message.
Run SerpAPI searches via SerpAPI's MCP server using mcporter. Use when the user asks to search the web with SerpAPI/SerpAPI MCP, wants SerpAPI inside Clawdbot, or to use the /serp command.
--- name: serpapi-mcp-9c244aecc6dd description: Run SerpAPI searches via SerpAPI's MCP server using mcporter. Use when the user asks to search the web with SerpAPI/SerpAPI MCP, wants SerpAPI inside Clawdbot, or to use the /serp command. --- # serpapi-mcp A wrapper skill that calls **SerpAPI’s MCP server** (Model Context Protocol) via the **`mcporter`** CLI. ## What this skill provides - Runs SerpAPI searches from OpenClaw via MCP (HTTP). - Returns the full SerpAPI JSON to stdout (the primary contract). - Optionally persists each query + response into **Airtable**: - raw JSON (full SerpAPI payload) - a structured “summary” (Top 10 organics, PAA/Related questions, videos, images, counts, flags) ## /serp usage Treat this skill as providing the `/serp` command. Syntax: - `/serp <query>` - `/serp <query> [engine] [num] [mode]` Defaults: - `engine=google_light` - `num=5` - `mode=compact` Notes: - If you want SERP features such as **People also ask (PAA)**, video packs, knowledge graph, etc., prefer: - `engine=google` - `mode=complete` Examples: - `/serp site:cnmv.es "educación financiera"` - `/serp "AAPL stock" google 3 compact` - `/serp "mortgage pay off vs invest" google 10 complete` ## How it works Main script (prints JSON to stdout): - `skills/serpapi-mcp/scripts/serp.sh "<query>" [engine] [num] [mode]` It calls this MCP tool endpoint: - `https://mcp.serpapi.com/$SERPAPI_API_KEY/mcp.search` Optional Airtable logger: - `skills/serpapi-mcp/scripts/airtable_log.mjs` ## Requirements ### 1) MCP client (`mcporter`) This skill requires **`mcporter`** installed on the host. Install: - `npm install -g mcporter` Verify: - `mcporter --help` ### 2) SerpAPI key(s) + failover You can configure either a single key or a failover pool: - Single key: - `SERPAPI_API_KEY` - Failover pool (comma-separated, tried in order): - `SERPAPI_API_KEYS` Recommended placement (any of these is supported): - Skill-scoped: - `skills.entries.serpapi-mcp.env.SERPAPI_API_KEY` - `skills.entries.serpapi-mcp.env.SERPAPI_API_KEYS` - Global: - `env.vars.SERPAPI_API_KEY` - `env.vars.SERPAPI_API_KEYS` Failover behavior: - The script retries with the next key on common **quota / auth / rate-limit** errors (e.g. 429/401/403, “quota exceeded”, “rate limit”). - For other errors (e.g. malformed request), it stops early and returns the error. ## Optional: store searches/results in Airtable ### Enable / disable Enable logging: - `SERP_LOG_AIRTABLE=1` (or `true`) You can set this globally in the Gateway env (always on) or per-run when executing the script. ### Airtable configuration (Gateway env) Set these environment variables (do not store secrets in the repo/workspace): - `AIRTABLE_TOKEN` (Airtable Personal Access Token) - `AIRTABLE_BASE_ID` - `AIRTABLE_TABLE` (table name **or** table id) ### Write behavior & compatibility - Airtable does **not** auto-create fields. - The logger is **schema-aware**: - It reads the table schema via Airtable Metadata API. - It only writes fields that already exist in your table (by name). - It coerces values to the Airtable field type when possible (checkbox/number/text/select/date). ### Airtable table schema (recommended) Field names must match exactly. Core: - `Query` → Single line text - `Engine` → Single line text - `Num` → Number (integer) - `Mode` → Single line text - `CreatedAt` → Date/Time Provenance: - `SerpApiSearchId` → Single line text - `SerpApiJsonEndpoint` → URL - `GoogleUrl` → URL Raw payload: - `ResultJson` → Long text - `ResultJsonTruncatedFlag` → Checkbox - (Back-compat: the logger also supports `ResultJsonTruncated` if you prefer that name.) Structured summary: - `SummaryJson` → Long text - `SummaryJsonTruncated` → Checkbox - `OrganicTop10Json` → Long text - `RelatedQuestionsTop10Json` → Long text - `ShortVideosTop10Json` → Long text - `VideosTop10Json` → Long text - `ImagesTop10Json` → Long text Flags + counts: - `HasAiOverview` → Checkbox - `HasAnswerBox` → Checkbox - `HasKnowledgeGraph` → Checkbox - `OrganicCount` → Number (integer) - `RelatedQuestionsCount` → Number (integer) - `ShortVideosCount` → Number (integer) - `VideosCount` → Number (integer) - `ImagesCount` → Number (integer) ### Airtable size limits Airtable has per-cell size limits. The logger truncates JSON strings if needed: - `AIRTABLE_MAX_JSON_CHARS` (default: 90000) - `AIRTABLE_MAX_SUMMARY_CHARS` (default: 90000) ## Output Returns SerpAPI JSON. Depending on engine and what Google shows, the payload may include keys like: - `organic_results` - `short_videos` / `videos_results` - `images_results` - `related_questions` - `knowledge_graph` - `answer_box` / `ai_overview` - ads blocks (`top_ads`, `bottom_ads`, etc.)--- name: serpapi-mcp-9c244aecc6dd description: Run SerpAPI searches via SerpAPI's MCP server using mcporter. Use when the user asks to search the web with SerpAPI/SerpAPI MCP, wants SerpAPI inside Clawdbot, or to use the /serp command. --- # serpapi-mcp A wrapper skill that calls **SerpAPI’s MCP server** (Model Context Protocol) via the **`mcporter`** CLI. ## What this skill provides - Runs SerpAPI searches from OpenClaw via MCP (HTTP). - Returns the full SerpAPI JSON to stdout (the primary contract). - Optionally persists each query + response into **Airtable**: - raw JSON (full SerpAPI payload) - a structured “summary” (Top 10 organics, PAA/Related questions, videos, images, counts, flags) ## /serp usage Treat this skill as providing the `/serp` command. Syntax: - `/serp <query>` - `/serp <query> [engine] [num] [mode]` Defaults: - `engine=google_light` - `num=5` - `mode=compact` Notes: - If you want SERP features such as **People also ask (PAA)**, video packs, knowledge graph, etc., prefer: - `engine=google` - `mode=complete` Examples: - `/serp site:cnmv.es "educación financiera"` - `/serp "AAPL stock" google 3 compact` - `/serp "mortgage pay off vs invest" google 10 complete` ## How it works Main script (prints JSON to stdout): - `skills/serpapi-mcp/scripts/serp.sh "<query>" [engine] [num] [mode]` It calls this MCP tool endpoint: - `https://mcp.serpapi.com/$SERPAPI_API_KEY/mcp.search` Optional Airtable logger: - `skills/serpapi-mcp/scripts/airtable_log.mjs` ## Requirements ### 1) MCP client (`mcporter`) This skill requires **`mcporter`** installed on the host. Install: - `npm install -g mcporter` Verify: - `mcporter --help` ### 2) SerpAPI key(s) + failover You can configure either a single key or a failover pool: - Single key: - `SERPAPI_API_KEY` - Failover pool (comma-separated, tried in order): - `SERPAPI_API_KEYS` Recommended placement (any of these is supported): - Skill-scoped: - `skills.entries.serpapi-mcp.env.SERPAPI_API_KEY` - `skills.entries.serpapi-mcp.env.SERPAPI_API_KEYS` - Global: - `env.vars.SERPAPI_API_KEY` - `env.vars.SERPAPI_API_KEYS` Failover behavior: - The script retries with the next key on common **quota / auth / rate-limit** errors (e.g. 429/401/403, “quota exceeded”, “rate limit”). - For other errors (e.g. malformed request), it stops early and returns the error. ## Optional: store searches/results in Airtable ### Enable / disable Enable logging: - `SERP_LOG_AIRTABLE=1` (or `true`) You can set this globally in the Gateway env (always on) or per-run when executing the script. ### Airtable configuration (Gateway env) Set these environment variables (do not store secrets in the repo/workspace): - `AIRTABLE_TOKEN` (Airtable Personal Access Token) - `AIRTABLE_BASE_ID` - `AIRTABLE_TABLE` (table name **or** table id) ### Write behavior & compatibility - Airtable does **not** auto-create fields. - The logger is **schema-aware**: - It reads the table schema via Airtable Metadata API. - It only writes fields that already exist in your table (by name). - It coerces values to the Airtable field type when possible (checkbox/number/text/select/date). ### Airtable table schema (recommended) Field names must match exactly. Core: - `Query` → Single line text - `Engine` → Single line text - `Num` → Number (integer) - `Mode` → Single line text - `CreatedAt` → Date/Time Provenance: - `SerpApiSearchId` → Single line text - `SerpApiJsonEndpoint` → URL - `GoogleUrl` → URL Raw payload: - `ResultJson` → Long text - `ResultJsonTruncatedFlag` → Checkbox - (Back-compat: the logger also supports `ResultJsonTruncated` if you prefer that name.) Structured summary: - `SummaryJson` → Long text - `SummaryJsonTruncated` → Checkbox - `OrganicTop10Json` → Long text - `RelatedQuestionsTop10Json` → Long text - `ShortVideosTop10Json` → Long text - `VideosTop10Json` → Long text - `ImagesTop10Json` → Long text Flags + counts: - `HasAiOverview` → Checkbox - `HasAnswerBox` → Checkbox - `HasKnowledgeGraph` → Checkbox - `OrganicCount` → Number (integer) - `RelatedQuestionsCount` → Number (integer) - `ShortVideosCount` → Number (integer) - `VideosCount` → Number (integer) - `ImagesCount` → Number (integer) ### Airtable size limits Airtable has per-cell size limits. The logger truncates JSON strings if needed: - `AIRTABLE_MAX_JSON_CHARS` (default: 90000) - `AIRTABLE_MAX_SUMMARY_CHARS` (default: 90000) ## Output Returns SerpAPI JSON. Depending on engine and what Google shows, the payload may include keys like: - `organic_results` - `short_videos` / `videos_results` - `images_results` - `related_questions` - `knowledge_graph` - `answer_box` / `ai_overview` - ads blocks (`top_ads`, `bottom_ads`, etc.)
Imported from the listed AgentSkill.sh category. Complete supplied package files and notices are retained. Includes source scripts; review them before use. Required external tools and credentials are not installed. No bundled code was executed during import. Individual agent task behavior has not been tested.