Tools
A set of executable operations exposed by an MCP Server. The model decides autonomously whether to invoke them; they carry side effects and require authorization from the user or the Host.
Core Facts
Timeline (last 90 days)
After Tools parsing is moved up, a logging Advisor can monitor the information passed by Tools, solving the black-box problem in older versions where upper-layer Advisors could not see the content passed by Tools.
Among the 11 benchmarks published by DeepSeek, 1017 surpasses the previous generation on all of them: Tools rose from 36.8 to 42.6, DeepSWE from 74.2 to 77.0, ProgramBench from 20.3 to 29.1, and TerminalBench 4.0 from 31.2 to 42.3.
MCP's classification standard for Tools is not based on whether they read or write, but on whether they need to execute a piece of logic based on input parameters
A single tool should ideally do only one clearly defined thing; the simpler the tool, the lower the probability of call errors, the easier the troubleshooting, and the more independently testable it is.
In an Agent architecture, the Planner layer is responsible for breaking down vague requirements into executable steps and is the true decision-maker determining what to do next, while Tools are merely executors.
Real pain points in enterprise-level AI application development include multi-layer memory architecture, difficulty selecting Tools, MCP authorization, and improving RAG retrieval accuracy.
Tools calling (Function Calling) allows a model to actively trigger predefined functions during a conversation and incorporate the results into its response, and was released by large model vendors earlier than MCP.
Tools are the capability interfaces through which agents take action: read is used to analyze problems, edit or write is used to create and modify code, and bash is used to run tests.
Tools (Function Calling) is the method by which large models communicate with one's own system, achieved by registering business method names, descriptions, and parameter structures with the large model to enable invocation.
Tools（Function Calling）是让大模型与自己内部系统进行通信的方式，基础大模型本身不具备调用外部API的能力
1 more timeline events
All Facts (12)
Tools (Function Calling) is the method by which large models communicate with one's own system, achieved by registering business method names, descriptions, and parameter structures with the large model to enable invocation.
65%UnverifiedTools选择困难问题在工具数量超过20个后尤为突出
60%UnverifiedAfter Tools parsing is moved up, a logging Advisor can monitor the information passed by Tools, solving the black-box problem in older versions where upper-layer Advisors could not see the content passed by Tools.
50%UnverifiedAmong the 11 benchmarks published by DeepSeek, 1017 surpasses the previous generation on all of them: Tools rose from 36.8 to 42.6, DeepSWE from 74.2 to 77.0, ProgramBench from 20.3 to 29.1, and TerminalBench 4.0 from 31.2 to 42.3.
50%UnverifiedMCP's classification standard for Tools is not based on whether they read or write, but on whether they need to execute a piece of logic based on input parameters
50%UnverifiedA single tool should ideally do only one clearly defined thing; the simpler the tool, the lower the probability of call errors, the easier the troubleshooting, and the more independently testable it is.
50%UnverifiedIn an Agent architecture, the Planner layer is responsible for breaking down vague requirements into executable steps and is the true decision-maker determining what to do next, while Tools are merely executors.
50%UnverifiedReal pain points in enterprise-level AI application development include multi-layer memory architecture, difficulty selecting Tools, MCP authorization, and improving RAG retrieval accuracy.
50%UnverifiedTools calling (Function Calling) allows a model to actively trigger predefined functions during a conversation and incorporate the results into its response, and was released by large model vendors earlier than MCP.
50%UnverifiedTools are the capability interfaces through which agents take action: read is used to analyze problems, edit or write is used to create and modify code, and bash is used to run tests.
50%UnverifiedTools（Function Calling）是让大模型与自己内部系统进行通信的方式，基础大模型本身不具备调用外部API的能力
50%UnverifiedTools面向内部系统的业务方法，MCP面向外部现成的第三方服务
50%