MCP-Cosmos: World Model-Augmented Agents for Complex Task Execution in MCP Environments

arXiv 2026 57.6 method, system

TLDR

MCP-Cosmos integrates generative world models into MCP to enable predictive task automation, improving agent performance on benchmark tasks.

Reasoning

The paper addresses a key gap between planning and execution by introducing a framework that uses world models for state simulation and plan refinement. Strengths include a novel BYOWM strategy and empirical evaluation on 20+ tasks. Weaknesses are limited details on world models and tasks, and lack of real-world validation.

Read-first score

Read-first score 57.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 45.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,evaluation,experiment,metric

Citation impact 18%

75.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.75708111

Topical relevance 29%

64.3

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 315.

Keyword Scores

world model

generative world model

world simulator

world dynamics prediction

interactive world model

model-based reinforcement learning world model

video world model

Deep Analysis

Innovations

Infusing generative World Models into the MCP ecosystem for predictive task automation
Bring Your Own World Model (BYOWM) strategy allowing agents to simulate state transitions and refine plans in latent space before execution
New metrics such as Execution Quality to evaluate world model effectiveness

Methodology

MCP-Cosmos is a framework that unifies MCP, World Model, and Agent technologies. It employs two agent strategies (ReAct and SPIRAL) with 2 planning models and 3 representative world models, evaluated over 20+ MCP-Bench tasks. The evaluation measures environment interaction KPIs including tool success rate and tool parameter accuracy.

Key Results

The framework showed improvements in tool success rate and tool parameter accuracy. The new Execution Quality metric provided insights into the effectiveness of world models compared to baselines.

Resources

Paper