Related Servers

The tool set contains many overlapping tools for similar operations. For example, there are multiple node-creation tools (create_workspace_node_from_scratch, create_workspace_node_from_predecessor, create_node_from_external_schema, create_pipeline_from_plan, create_pipeline_from_sql, build_pipeline_from_intent, pipeline_workshop) and multiple update tools (set_workspace_node, update_workspace_node, replace_workspace_node_columns, apply_join_condition, convert_join_to_aggregation) whose boundaries require very detailed descriptions to disambiguate. Similarly, run-related tools (start_run, run_and_wait, retry_run, retry_and_wait, get_run, get_run_details, run_status, get_run_results) heavily overlap in purpose.

Naming is mostly snake_case verb_noun (list_workspaces, create_environment, get_run), but there is inconsistency: some tools use different verb forms (set_env_role vs delete_env_role; create vs update vs set), some omit the noun (run_status vs get_run_status), and there are several variants that break the pattern (coa_doctor, coa_bootstrap_workspaces, pipeline_workshop_open, plan_pipeline). The coa_* prefix family is consistent internally but the overall set mixes generic verbs (get, list, create, update, delete, set, run, start, retry, build, plan) without a clear convention about which verb to use for which operation.

115 tools is an extreme count for an MCP server. Even for a full Coalesce platform surface (projects, environments, workspaces, nodes, jobs, runs, roles, git accounts, pipelines, coa CLI), the count is bloated by many composite/derived tools (get_run_details, get_environment_overview, run_and_wait, retry_and_wait) and near-duplicate variants. This will cause selection overhead and confusion for an agent.

Within its broad domain, the tool surface is remarkably complete: CRUD for projects, environments, workspaces, node mutations, role management, git accounts, jobs, runs, subgraphs, plus pipeline planning/creation, lineage analysis, caching, and a full coa CLI wrapper. There are no obvious dead ends — even setup diagnosis and skill personalization are covered.