[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86537-en":3,"doc-seo-86537-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},86537,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",6,"Technology","ToolAtlas Learning Once Reusing Everywhere with Tool Side Memory","Large language model (LLM) agents increasingly depend on externally hosted tools supplied by shared providers and called by diverse downstream agents. Prior work improves tool use mainly on the agent side via parameter updates, prompt tuning, or agent memory, which restricts reuse to behaviors seen in past tasks. ToolAtlas proposes a provider-side, graph-based persistent tool memory that captures capabilities, failure boundaries, and cross-tool compositions via execution-verified probing, queried through adaptive traversal. Across MCP benchmarks, it improves pass@1 and pass@4 and transfers without retraining or task-time exploration.","ToolAtlas: Learning Once, Reusing Everywhere with Tool-Side Memory  \nYue Fang♠ * , Zhibang Yang♠ * , Fangkai Yang♣†, Xiaoting Qin♣†, Liqun Li♣†, Qingwei Lin♣ , Saravan Rajmohan♣ , Dongmei Zhang♣  \n♠ School of Computer Science, Peking University, Beijing, China  \n♣Microsoft  \narXiv :2607 . 1 1 126v 1 [ cs .LG] 13 Jul 2026  \nAbstract  \nLarge language model (LLM) agents increasingly rely on external tools served by shared providers and accessed by heterogeneous downstream agents. Existing approaches improve tool use on the agent side through parameter updates, prompt refinement, or agent-side memory, making tool knowledge difficult to share and limited to behaviors observed in past tasks. We argue that reusable tool knowledge should instead be maintained by the tool provider. We introduce ToolAtlas, a graph-based framework that builds a persistent provider-side tool memory of tool capabilities, failure boundaries, and cross-tool compositions through executionverified probing. At inference time, agents query the tool memory via adaptive graph traversal. Across two MCP-based benchmarks spanning eight services, ToolAtlas outperforms existing tool-side optimization and agentside memory baselines by up to 21.61% in pass@1 and 18.61% in pass@4 . The same tool memory also transfers across environment instances and agent frameworks without retraining or task-time exploration, yielding up to 24.16%/16.22% and 17.49%/14.27% relative gains in pass@1/pass@4, respectively. Ablation studies show that these gains arise from combining tool-centered memory organization with capability-guided execution probing. These results establish provider-side tool memory as an effective and reusable paradigm for tool servers. Our code is in: [https://github](https://github). com/PuppyKnightUniversity/ToolAtlas.  \n1 Introduction  \nTool use has become a defining capability of modern large language model (LLM) agents, enabling them to invoke external services and data sources beyond their parametric knowledge. Recent work shows that such tool use now spans large, realistic ecosystems, from comprehensive tool-augmented  \n*These authors contributed equally.  \n†Corresponding author.  \nFigure 1: Two structural limitations of agent-side tool memory in shared tool server settings.  \nbenchmarks to thousands of real-world APIs (Liet al., 2023 ; Patil et al., 2024 ; Qin et al., 2024) . Under the Model Context Protocol (MCP), this ecosystem is increasingly deployed in a server-centric form: tool providers expose capabilities through MCP servers, while heterogeneous agents consume them across products and environments (Model Context Protocol, 2026 ; OpenAI, 2026) . Official provider-hosted servers from GitHub and Stripe further reflect this trend (GitHub, 2026 ; Stripe, 2026) . In this setting, the same tool server may be called by many downstream agents with different prompts, action interfaces, and runtime environments.  \nDespite this shared-server setting, most prior work still improves tool use on the agent side. Agent-side methods internalize tool-use behavior into model parameters (Schick et al., 2023 ; Qin et al., 2024 ; Liu et al., 2025b), create reusable skills or tools (Wang et al., 2024a ; Cai et al., 2024 ; Qian et al., 2023 ; Yuan et al., 2024), or accumulate experience in agent-side memory (Zhao et al., 2024 ; Ouyang et al., 2025 ; Xu et al., 2025 ; Zhang et al., 2025a ; Xiao et al., 2025) . Existing tool-side efforts refine tool descriptions, documentation, retrieval, or catalog organization (Yuan et al., 2025 ; Fanget al., 2025 ; Wu et al., 2025a ; Qu et al., 2025 ; Liu et al., 2025a) . However, existing tool-side opti-  \nmization mainly improves how tools are presented or selected, while agent-side memory leaves knowledge inside the agent that produced it, rather than making it reusable across agents that use same tool.  \nThis ownership mismatch matters in sharedserver deployments. When many agents access one MCP server, each agent must rediscover behavio","cbCaikkLpCbA21JC","https://ap.wps.com/l/cbCaikkLpCbA21JC","pdf",1910052,1,23,"English","en",105,"# Introduction\n## Shared tool-server setting under MCP\n## Limitations of agent-side tool memory\n## Challenges: generalization dilemma and capability blindness\n## Proposed solution: provider-side tool memory graph","[{\"question\":\"What problem does ToolAtlas target in shared MCP tool-server deployments?\",\"answer\":\"It targets the mismatch where tool knowledge is stored on the agent side, forcing every agent to rediscover tool behaviors and failure modes instead of reusing provider-maintained knowledge.\"},{\"question\":\"How does ToolAtlas build reusable tool knowledge?\",\"answer\":\"ToolAtlas constructs a provider-side, graph-based tool memory that records capabilities, failure boundaries, and cross-tool compositions using execution-verified probing rather than passive trajectory logging.\"},{\"question\":\"How is the tool memory used at inference time?\",\"answer\":\"At inference time, agents query the provider-side tool memory using adaptive graph traversal to find relevant tool capabilities and compositions.\"}]",1784212484,58,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"toolatlas-learning-once-reusing-everywhere-with-tool-side-memory","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/technology/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/toolatlas-learning-once-reusing-everywhere-with-tool-side-memory/86537/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does ToolAtlas target in shared MCP tool-server deployments?","Question",{"text":75,"@type":76},"It targets the mismatch where tool knowledge is stored on the agent side, forcing every agent to rediscover tool behaviors and failure modes instead of reusing provider-maintained knowledge.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does ToolAtlas build reusable tool knowledge?",{"text":80,"@type":76},"ToolAtlas constructs a provider-side, graph-based tool memory that records capabilities, failure boundaries, and cross-tool compositions using execution-verified probing rather than passive trajectory logging.",{"name":82,"@type":73,"acceptedAnswer":83},"How is the tool memory used at inference time?",{"text":84,"@type":76},"At inference time, agents query the provider-side tool memory using adaptive graph traversal to find relevant tool capabilities and compositions.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,113,118,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":111,"slug":112},50,"technology",{"id":114,"doc_module":4,"doc_module_name":45,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":119,"doc_module":4,"doc_module_name":45,"category_name":120,"show_sort_weight":121,"slug":122},8,"Research & Report",30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]