Definition
Tool affordance is the signal an LLM agent uses to decide whether to call a given tool: the tool's name, description text, prior success rate (in agent runtimes that track this), and contextual cues from the user's prompt. Better-named, less-overlapping tools produce better routing decisions.
In Depth
When 5 search tools have descriptions like 'search the web', 'lookup online', 'find information', the LLM picks differently per prompt phrasing — a coin flip. The fix is two-pronged: (1) consolidate to one tool per category (Scavio MCP replaces 5-8 narrow search tools); (2) make remaining tool names + descriptions clear and orthogonal ('scavio.search for web SERP', 'scavio.reddit_search for community signal', 'scavio.youtube_search for video content'). Routing accuracy correlates with affordance clarity. Agent runtimes that track per-tool success rates can up-weight high-affordance tools dynamically; most don't, so the manual fix matters more.
Example Usage
Agent has 'web_search', 'google_lookup', 'serp_query', 'general_search' all with similar descriptions. LLM picks inconsistently. Consolidating to 'scavio.search' (one tool, clear affordance) routes 95%+ of queries to the right tool with no ambiguity.
Platforms
Tool Affordance (Agent) is relevant across the following platforms, all accessible through Scavio's unified API:
Related Terms
Agent Routing Layer
An agent routing layer is the component (often custom code or a framework primitive like LangGraph) that sits between th...
Agent Skill Bloat
Agent skill bloat is the per-message token-cost compounding effect of leaving 50-100 default skills loaded in an agent's...
MCP Tool Naming Convention
MCP tool naming convention is the discipline of naming Model Context Protocol tools clearly and orthogonally (e.g., scav...