Related Servers

Several tools are near-duplicates distinguished only by the underlying API or auth mechanism: cs_list_environments, cs_env_list, and cs_admin_list_environments all list environments; cs_pull/cs_pull_solution and cs_clone_agent/cs_clone_solution overlap; and five separate flow-run tools (cs_list_flow_runs, cs_get_flow_run, cs_explain_flow_run, cs_compare_flow_runs, cs_analyze_flow_health) require careful reading to pick correctly. The descriptions do attempt to differentiate, and subdomain prefixes help, but at 142 tools the selection risk is high.

The cs_ prefix and a mostly verb_noun pattern (list_*, create_*, update_*, add_*) create a recognizable system, with consistent subdomain prefixes like cs_admin_ and cs_env_. However, there are many bare-verb outliers (cs_init, cs_pull, cs_push, cs_pack, cs_chat, cs_guide, cs_validate), noun-first inversions (cs_env_list alongside cs_list_environments, cs_env_fetch, cs_auth_who), and noun-noun names like cs_job_status, so the pattern is not reliably predictable.

142 tools consuming roughly 50k tokens of schema is far beyond what an agent can use effectively, and the server itself ships cs_set_tool_preset specifically to hide most tools because the full list degrades model performance. The breadth of the domain (agent authoring, Power Automate, Dataverse, solutions, tenant admin, Graph) justifies a large surface, but this is effectively several servers' worth of tools bundled into one, and most sessions need only a small subset.

For its stated scope, coverage is unusually thorough: agent CRUD, topics, knowledge, tools, flows, variables, triggers, evaluations, test sets, transcripts, drift detection, solutions, connections, auth, tenant admin, and org-agent governance are all represented with no obvious dead ends. Minor gaps remain — there is no way to chat with the draft agent directly (cs_chat targets only the published one), and some connector authorizations are portal-only — but these are documented rather than silent omissions.