How do I install the Run Tool skill?

Run npx skillmds@latest add mosofin/run-tool in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Run Tool safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Run Tool?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Run Tool free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Run Tool?

MosoFin (@mosofin) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Run Tool by mosofin — Agent Skill

Name: Run Tool
Author: MosoFin

Run Tool

Run one specific MosoFin catalog operation against a connected company file (maps to the invoke_datasource_api_tool MCP tool). Use when the user names an exact operation to execute (e.g. run get_profit_and_loss, call search_invoices) rather than asking an open data question.

MosoFin 3c1eae1 2.2 KB Updated Sep 17, 2026

File contents

Run a catalog operation

Wraps the server's invoke_datasource_api_tool — the only data-egress tool. Full spec: docs/mcp-tool-spec.md §3.4. For open-ended questions ("how did July look?"), prefer /mosofin:query-workspace.

Steps

Confirm workspace (list_workspaces) and, when several company files are connected, the company by display_name (get_agent_datasources) if not already confirmed in this chat.
Verify the requested tool_name exists in the live catalog (get_datasource_tools) when unsure — on UNKNOWN_TOOL, pick from the valid-names list; never guess.
Build params:
- Period reports and searches: concrete start_date + end_date (YYYY-MM-DD) — resolve relative phrases first; ask if ambiguous.
- As-of reports: optional report_date (default today).
- Get-by-id: id.
Invoke with datasource, tool_name, params, workspace_id, and data_source_id (both on every call and retry — the server is stateless). Independent invokes may batch in parallel.
Handle results and errors:
- Unknown tool / MosoFin tools missing → never say "refresh/reconnect". Give ChatGPT/Codex setup: Settings → Apps & Connectors → create MosoFin at https://mcp.mosofin.com/mcp with OAuth → sign in → new chat with MosoFin on. No data.
- pagination.has_more → re-invoke with params.offset = pagination.next_offset.
- approval_required → ask the user, retry the same invoke with approved=true. Never pre-set it.
- entity_required → ask which company from entities[]; retry.
- datasource_not_active → give the user reconnect_url; no data.
- invalid_params → fix the named field; retry.
- tool_policy_disabled → explain; do not retry.
Check mock (live vs fixture) and end with a Data sources line (datasource + tool + fetched_at from provenance). Never fabricate figures on failure.

---name: run-tooldescription: Run one specific MosoFin catalog operation against a connected company file (maps to the invoke_datasource_api_tool MCP tool). Use when the user names an exact operation to execute (e.g. run get_profit_and_loss, call search_invoices) rather than asking an open data question.---# Run a catalog operationWraps the server's `invoke_datasource_api_tool` — the **only data-egresstool**. Full spec: `docs/mcp-tool-spec.md` §3.4. For open-ended questions("how did July look?"), prefer `/mosofin:query-workspace`.## Steps1. Confirm workspace (`list_workspaces`) and, when several company files are connected, the company by `display_name` (`get_agent_datasources`) if not already confirmed in this chat.2. Verify the requested `tool_name` exists in the live catalog (`get_datasource_tools`) when unsure — on `UNKNOWN_TOOL`, pick from the valid-names list; never guess.3. Build `params`: - Period reports and searches: concrete `start_date` + `end_date` (`YYYY-MM-DD`) — resolve relative phrases first; ask if ambiguous. - As-of reports: optional `report_date` (default today). - Get-by-id: `id`.4. Invoke with `datasource`, `tool_name`, `params`, `workspace_id`, and `data_source_id` (both on **every** call and retry — the server is stateless). Independent invokes may batch in parallel.5. Handle results and errors: - `Unknown tool` / MosoFin tools missing → never say "refresh/reconnect". Give ChatGPT/Codex setup: Settings → Apps & Connectors → create MosoFin at `https://mcp.mosofin.com/mcp` with OAuth → sign in → new chat with MosoFin on. No data. - `pagination.has_more` → re-invoke with `params.offset = pagination.next_offset`. - `approval_required` → ask the user, retry the **same** invoke with `approved=true`. Never pre-set it. - `entity_required` → ask which company from `entities[]`; retry. - `datasource_not_active` → give the user `reconnect_url`; no data. - `invalid_params` → fix the named field; retry. - `tool_policy_disabled` → explain; do not retry.6. Check `mock` (live vs fixture) and end with a **Data sources** line (datasource + tool + `fetched_at` from `provenance`). Never fabricate figures on failure.

Frequently asked questions