Preprint states that multiple LLM clients misselect counterfeit tools in MCP
Start here
A preprint posted to arXiv on November 16 studies the tool provider misbinding problem in MCP agents: tool metadata is not bound to a verified provider identity, so an attacker only needs to spin up another MCP server that publishes tools with self-chosen metadata, without compromising the host, the LLM, a legitimate server, or the network channel, to make requests intended for a legitimate provider resolve to an attacker-controlled tool. The authors evaluated five LLM-based clients in a controlled MCP environment and found that misbinding was strongest when exact name collisions, task-aligned descriptions, and compatible schemas reinforced one another; authority-enhanced descriptions that do not rely on name collisions, combined with compatible schemas, could also lead to execution by the wrong provider or to unresolvable provider ambiguity. Under conditions where tool names were exactly identical, descriptions were copied verbatim, and schemas were compatible, GPT-5.4, Qwen3.7 Max, and DeepSeek V3 all selected the attacker's tool in all 50 trials, Gemini 2.5 Pro did so in 84%, and Claude Sonnet 5 in 6%; Claude Sonnet 5 requested clarification first in 94% of trials. The authors propose provider-bound tool identity with signed manifests and stable provider identifiers as a remediation direction, but the paper did not evaluate these mechanisms. The experiments covered only one benign dataset retrieval task, and this is a preprint that has not undergone peer review.
Generated by AI from reports · Updated 16 hr ago
Attention and developments
Attention, hourlyA report
Now 44, +20% in 24 hours·Peak 47 (Nov 19 13:00)·1 report in this range, 1 earlier
Progress timeline
- TechTimesSelectedPreprint: when name, description and schema all match, three models including GPT-5.4 chose the impersonating MCP tool in all 50 trials
A preprint posted to arXiv on November 16 shows that standing up a separate MCP server with impersonating tool metadata, without compromising the host, the model or the legitimate server, can make models hand requests to the attacker's tool. Under the condition that tool names were exactly identical, descriptions were copied verbatim and schemas were compatible, GPT-5.4, Qwen3.7 Max and DeepSeek V3 all selected the attacker's tool in all 50 trials under that condition, Gemini 2.5 Pro was 84%, and Claude Sonnet 5 was 6%. Under this condition, Claude Sonnet 5 first asked for clarification in 94% of trials, while the other three models never asked. The authors call this failure tool-provider misbinding and propose provider-bound tool identity with signed manifests and stable provider identifiers as a fix, but the paper did not evaluate these mechanisms. The experiment covered only one benign dataset-acquisition task and is a preprint that has not undergone peer review.
- arXiv: 其他 AI 安全论文When tool names are not identity: measuring provider misbinding in MCP-based AI agents
The paper studies the tool provider misbinding problem in MCP agents: the metadata used to discover and select tools is not itself bound to a verified provider identity. An attacker does not need to compromise the host, the LLM, a legitimate server, or the network channel; they only need to run another MCP server that publishes tools with metadata of their choosing, and requests that should go to a legitimate provider can then resolve to an attacker-controlled tool. The authors formalize this failure as an attacker-controlled tool selection optimization problem and evaluate five LLM-based clients in a controlled MCP environment. The results show interaction-driven misbinding: failures are strongest when exact name conflicts, task-aligned descriptions, and compatible schemas reinforce one another; authority-enhanced descriptions that do not rely on name conflicts, together with compatible schemas, can also lead to execution by the wrong provider or to unresolvable provider ambiguity. Different models expose the same binding weakness in different ways, including attacker-controlled provider selection, requests for clarification, and unstable or inconsistent provider resolution.
Related discussion
- X@Cy5una tool name isn't an identity. this paper adds one normal mcp server with a same-named tool and matching schema, and c…(opens the original post in a new tab)
1 likes·Nov 19 16:45Collect
- X@FSFGWhen Tool Names Are Not Identities: Measuring Provider Misbinding in MCP-Based AI Agents Zeynab Anbiaee, Sajjad Dadkha…(opens the original post in a new tab)
- X@bryanschwabTool names are not identities. 5 LLM clients, one extra MCP server advertising look-alike tools. A matching name, desc…(opens the original post in a new tab)
14 likes·Nov 19 11:42Collect
- X@xiaoanlabYour agent may already be handing its tools to an attacker — no hack required, just a same-named tool on another MCP se…(opens the original post in a new tab)
6 likes·Nov 19 11:42Collect
- X@gastronomyWhen Tool Names Are Not Identities: Measuring Provider Misbinding in MCP-Based AI Agents: Large language model (LLM) ap…(opens the original post in a new tab)
1 likes·Nov 19 11:42Collect
Attention trend
New sources per day
- Nov 17: 3 new sources (1 outlets, 2 accounts)
- Nov 18: 1 new sources (0 outlets, 1 accounts)
- Nov 19: 3 new sources (1 outlets, 2 accounts)
- Nov 20: 0 new sources (0 outlets, 0 accounts)