---

name: agentsbox description: >- Discovers and executes MCP tools on-demand via agentsbox_* tools. Use when you need to find the right tool, when a user asks to use a tool you don’t have in context, or when troubleshooting MCP/tool availability. license: MIT metadata: author: agentsbox version: "0.1"

agentsbox

Minimize context bloat by using a small, stable tool surface:

• agentsbox_search_bm25 (natural language)
• agentsbox_search_regex (pattern search)
• agentsbox_execute (run discovered tool)
• agentsbox_status (health/status)
• agentsbox_perf (performance)
• agentsbox_test (tool verification)

Workflow

1. Search first

   • If you don’t know the exact tool name: agentsbox_search_bm25.
   • If you know server prefix / partial name: agentsbox_search_regex.

2. Inspect schema

   • From search results, read schema and required fields.

3. Execute

   • Call agentsbox_execute({ toolId, arguments }).
   • toolId format: {serverName}_{toolName}.
   • arguments is a JSON string (or omit / {} when none required).

4. Troubleshoot

   • If tool not found: widen search, verify server prefix, try BM25.
   • If execution fails: run agentsbox_status({}) and include server diagnostics.

Examples

Search the web with tavily

1. Search:

agentsbox_search_bm25({ "text": "search the web for information", "limit": 5 })

2. Execute:

agentsbox_execute({
  "toolId": "tavily_tavily_search",
  "arguments": "{\"query\":\"latest AI news\",\"max_results\":5}"
})

Edge cases

• Empty or missing servers: agentsbox_status will show zero configured servers.
• Schema mismatch: prefer passing only required fields first.
• Server name/tool name ambiguity: use regex to list server tools: "server_.*".