Introducing Tool Recommendation: Save Up to 37% on Token Cost While Giving Your Agent the Whole Tool Catalog
Blog post from Arcade
Arcade.dev has introduced Tool Recommendation, a capability designed to help AI agents access large tool catalogs without loading every tool definition into their context, addressing the cost, context-window, and accuracy challenges that arise when enterprises connect agents to thousands of actions across systems such as Gmail, Salesforce, GitHub, and Jira. Instead of requiring an agent’s language model to choose among all available tools, the feature uses embedding and ranking models to retrieve a small set of relevant tools based on a plain-language request, then enables the agent to invoke them through Arcade’s MCP-compatible gateway. Arcade says this approach can reduce input tokens per session by up to 91% and modeled token costs by up to 37% compared with supplying a fixed list of 100 tools, while improving tool-retrieval performance in its benchmark comparisons, though its reported results include methodological qualifications. The feature operates within Arcade’s actions runtime, which applies user authorization, policy enforcement, data protection checks, auditing, and centralized observability to recommendations and tool calls. Tool Recommendation is available in Arcade Cloud and can also be deployed in customer-controlled cloud, on-premises, or air-gapped environments, with support for different models, frameworks, and MCP clients.
| Trend | Post Mentions | Total Month Mentions | Posts | Companies | MoM |
|---|---|---|---|---|---|
| MCP | 5 | 11,921 | 1,232 | 235 | +12% |
| Vector Search | 2 | 1,978 | 415 | 135 | +18% |
| LLM | 1 | 6,578 | 1,243 | 236 | +9% |
| Observability | 1 | 3,086 | 746 | 186 | +11% |
| OpenTelemetry | 1 | 860 | 177 | 65 | +21% |
| Secrets Management | 1 | 3,093 | 620 | 154 | +51% |
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.