SocialGPT

Most tools map to clearly distinct resources and actions, and the descriptions are unusually explicit about boundaries. A few clusters are still easy to mis-select—growth vs. account metrics, search vs. search_videos, and fetch vs. get_video—but an agent reading the descriptions should usually pick correctly.

The surface is overwhelmingly consistent snake_case verb_noun naming: list_accounts, get_creator, analyze_post, publish_post. The main deviations are the connector-spec names `search` and `fetch`, plus `server_info` and `whoami`, which prevent a perfect score.

26 tools is slightly heavy, but the server covers a broad life cycle: creator ingestion, video analysis, metrics, semantic search, uploading, publishing, and identity/scope inspection. While a few tools overlap at the edges, most have a distinct job, so the count is justifiable rather than bloated.

The tool surface covers the main workflows well: analyze creators/posts, read results, list videos, retrieve metrics, search semantic content, publish posts, handle uploads, and poll job status. Minor gaps exist, such as no way to list all creators in the analysis library or delete/refresh analyses/uploads, but agents can usually work around them.