{"paper":{"title":"ComplexMCP: Evaluation of LLM Agents in Dynamic, Interdependent, and Large-Scale Tool Sandbox","license":"http://creativecommons.org/licenses/by/4.0/","headline":"LLM agents reach under 60 percent success on interdependent tool tasks where humans hit 90 percent.","cross_cats":["cs.SE"],"primary_cat":"cs.AI","authors_text":"Hongyang Chen, Longyue Wang, Weihua Luo, Xue Yang, Yuanyang Li","submitted_at":"2026-05-11T16:20:51Z","abstract_excerpt":"Current LLM agents are proficient at calling isolated APIs but struggle with the \"last mile\" of commercial software automation. In real-world scenarios, tools are not independent; they are atomic, interdependent, and prone to environmental noise. We introduce $\\textbf{ComplexMCP}$, a benchmark designed to evaluate agents in these rigorous conditions. Built on the Model Context Protocol (MCP), $\\textbf{ComplexMCP}$ provides over 300 meticulously tested tools derived from 7 stateful sandboxes, ranging from office suites to financial systems. Unlike existing datasets, our benchmark utilizes a see"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"even top-tier models fail to exceed a 60% success rate, far trailing human performance 90%. Granular trajectory analysis identifies three fundamental bottlenecks: (1) tool retrieval saturation as action spaces scale; (2) over-confidence, where agents skip essential environment verifications; and (3) strategic defeatism, a tendency to rationalize failure rather than pursuing recovery.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The 7 stateful sandboxes and 300 tools derived from them sufficiently represent the interdependent, noisy conditions of real-world commercial software automation.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"ComplexMCP benchmark shows current LLM agents achieve at most 60% success on interdependent tool tasks versus 90% for humans, due to tool retrieval saturation, over-confidence, and strategic defeatism.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"LLM agents reach under 60 percent success on interdependent tool tasks where humans hit 90 percent.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"8057dbadd909034fb9f5b0fd4661a5d8d4f7110efb64319b3712927e8c22d70f"},"source":{"id":"2605.10787","kind":"arxiv","version":2},"verdict":{"id":"3b30be21-c64d-419d-8e6d-93bdba03522f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T04:37:47.107372Z","strongest_claim":"even top-tier models fail to exceed a 60% success rate, far trailing human performance 90%. Granular trajectory analysis identifies three fundamental bottlenecks: (1) tool retrieval saturation as action spaces scale; (2) over-confidence, where agents skip essential environment verifications; and (3) strategic defeatism, a tendency to rationalize failure rather than pursuing recovery.","one_line_summary":"ComplexMCP benchmark shows current LLM agents achieve at most 60% success on interdependent tool tasks versus 90% for humans, due to tool retrieval saturation, over-confidence, and strategic defeatism.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The 7 stateful sandboxes and 300 tools derived from them sufficiently represent the interdependent, noisy conditions of real-world commercial software automation.","pith_extraction_headline":"LLM agents reach under 60 percent success on interdependent tool tasks where humans hit 90 percent."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2605.10787/integrity.json","findings":[],"available":true,"detectors_run":[{"name":"claim_evidence","ran_at":"2026-05-20T05:22:00.382797Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"ai_meta_artifact","ran_at":"2026-05-19T14:35:51.574811Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_title_agreement","ran_at":"2026-05-19T10:31:17.789485Z","status":"completed","version":"1.0.0","findings_count":0},{"name":"doi_compliance","ran_at":"2026-05-19T08:58:52.075793Z","status":"completed","version":"1.0.0","findings_count":0}],"snapshot_sha256":"24642f0bbdb1fe908881af6cae8eeb19b84ccbd618d70c09890213eae89c513e"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}