MCP line jumping: attacks before the first tool use
Research showed malicious MCP tool descriptions can inject instructions before any tool runs, enabling code execution, data theft and approval bypass.
What happened
Malicious MCP servers inject via tool descriptions before any tool is invoked ('line jumping'), enabling code execution/exfil and approval bypass. Follow-up showed conversation-history theft paths and ANSI escape deception in tool UX.
How it happened
Read the flow as text & sources
A catalog reconstruction of the supplied source account. Unknown steps remain explicit; an edge describes the reported relationship between steps.
- 01 · Connection route is unspecified
The supplied record starts with a malicious MCP server available to the agent.
Source for step 1 - 02 · Tool descriptions enter model context
The attack content is ingested before the first tool use.
Hunt: Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.
Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.
Source for step 2 - 03 · Injected instructions jump the approval boundary
The model can be redirected without an initial explicit tool invocation.
Hunt: Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.
Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.
Source for step 3 - 04 · Demonstrated execution or exfiltration
The research describes approval bypass and conversation-history theft paths.
Hunt: Sensitive reads are followed by outbound transfer, a public artifact, or a tool call to a destination outside the approved workflow.
An outbound connection or high byte count alone is not exfiltration. Encrypted traffic may hide content, and bulk legitimate exports are common.
Source for step 4
Why it matters
Reported impact: The research describes approval bypass and conversation-history theft paths.
Check these source-linked conditions against your setup.
The research places malicious instructions in MCP descriptions before a tool is invoked.
Conditions to check- Review the connected server, model / client behavior and the permissions exercised by the demonstration.
Source-linked applicability guidance. Confirm current scope in the original report.
What you can do
Protection & prevention
Review server descriptions before connection and minimize tool and data access before approval.
TOFU pinning (mcp-context-protector); sanitize ANSI; never auto-trust tool metadata
Catalog defensive guidance based on the reported behavior. Confirm scope and current mitigations in the original source.
Detection queries
Draft query starters grouped by the attack stages where they apply.
Start with the logs you collect. Each hunt identifies its required visibility, field mapping and limits. Queries are catalog-authored drafts using reusable behavior profiles; test with known activity and your normal baseline before alerting.
02Tool descriptions enter model contextUnexpected agent tool useCustom schema · Draft
The attack content is ingested before the first tool use.
What to look for
Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.
Visibility: Defender-side logs where collected
Required logs
- Agent tool-call audit records with session / tool / destination
- Tool-definition and approval changes
- Endpoint process, file-access and network evidence where available
Coverage limit. Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.
Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.
Sigma
title: Hunt - Unexpected agent tool use
id: c709e384-0e11-479e-ab4d-3a06e2431e4b
status: experimental
description: Untrusted content or a changed tool definition precedes a file
read, command, or remote call outside the user's intended task.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
- Authorized activity matching this broad hunt; baseline and investigate
before alerting.
level: low
logsource:
category: application
product: openattacks_audit
definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,
Target and CorrelationId from the required source logs first.
detection:
selection:
EventType:
- agent.tool.call
- agent.tool.definition.change
condition: selection
KQL
// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.
// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).
// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.
OpenAttacksAudit_CL
| where TimeGenerated > ago(7d)
| where EventType in ("agent.tool.call", "agent.tool.definition.change")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated descSPL
sourcetype="openattacks:audit"
EventType IN ("agent.tool.call", "agent.tool.definition.change")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _timeInvestigation steps
- Join input provenance, tool descriptions and calls by session ID and compare the action to the approved task.
- Inspect child processes and destinations; record whether execution was in-process, local, or through a remote SaaS connector.
03Injected instructions jump the approval boundaryUnexpected agent tool useCustom schema · Draft
The model can be redirected without an initial explicit tool invocation.
What to look for
Untrusted content or a changed tool definition precedes a file read, command, or remote call outside the user's intended task.
Visibility: Defender-side logs where collected
Required logs
- Agent tool-call audit records with session / tool / destination
- Tool-definition and approval changes
- Endpoint process, file-access and network evidence where available
Coverage limit. Process logs cannot recover model intent. Remote connectors and in-process interpreters may produce no local child process. Prompt content alone is not a reliable signature.
Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.
Sigma
title: Hunt - Unexpected agent tool use
id: c709e384-0e11-479e-ab4d-3a06e2431e4b
status: experimental
description: Untrusted content or a changed tool definition precedes a file
read, command, or remote call outside the user's intended task.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
- Authorized activity matching this broad hunt; baseline and investigate
before alerting.
level: low
logsource:
category: application
product: openattacks_audit
definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,
Target and CorrelationId from the required source logs first.
detection:
selection:
EventType:
- agent.tool.call
- agent.tool.definition.change
condition: selection
KQL
// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.
// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).
// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.
OpenAttacksAudit_CL
| where TimeGenerated > ago(7d)
| where EventType in ("agent.tool.call", "agent.tool.definition.change")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated descSPL
sourcetype="openattacks:audit"
EventType IN ("agent.tool.call", "agent.tool.definition.change")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _timeInvestigation steps
- Join input provenance, tool descriptions and calls by session ID and compare the action to the approved task.
- Inspect child processes and destinations; record whether execution was in-process, local, or through a remote SaaS connector.
04Demonstrated execution or exfiltrationSensitive data → unusual destinationCustom schema · Draft
The research describes approval bypass and conversation-history theft paths.
What to look for
Sensitive reads are followed by outbound transfer, a public artifact, or a tool call to a destination outside the approved workflow.
Visibility: Defender-side logs where collected
Required logs
- Sensitive resource access / DLP records
- Proxy, egress and cloud upload events
- Agent / SaaS audit with correlation IDs
Coverage limit. An outbound connection or high byte count alone is not exfiltration. Encrypted traffic may hide content, and bulk legitimate exports are common.
Custom-schema candidates, not out-of-box detectors. Map your application / cloud / platform logs to EventType, Actor, Asset, Action, Target and CorrelationId. KQL uses a custom OpenAttacksAudit_CL table; SPL uses openattacks:audit. EventType values are proposed normalization names, not native vendor event names. Correlation and environment-specific alert logic remain to be built.
Sigma
title: Hunt - Sensitive data → unusual destination
id: 04b60968-3179-4b1f-ae13-6f57c2446077
status: experimental
description: Sensitive reads are followed by outbound transfer, a public
artifact, or a tool call to a destination outside the approved workflow.
author: MagicSword OpenAttacks catalog
date: 2026-10-23
falsepositives:
- Authorized activity matching this broad hunt; baseline and investigate
before alerting.
level: low
logsource:
category: application
product: openattacks_audit
definition: Custom normalized audit schema; map EventType, Actor, Asset, Action,
Target and CorrelationId from the required source logs first.
detection:
selection:
EventType:
- data.export
- agent.external.send
condition: selection
KQL
// SCHEMA TEMPLATE: create/map this custom table before use; not a built-in Sentinel table.
// OpenAttacksAudit_CL: TimeGenerated(datetime), EventType/Actor/Asset/Action/Target/CorrelationId(string).
// Selects candidates only. Apply the stage hypothesis and correlation steps before alerting.
OpenAttacksAudit_CL
| where TimeGenerated > ago(7d)
| where EventType in ("data.export", "agent.external.send")
| project TimeGenerated, EventType, Actor, Asset, Action, Target, CorrelationId
| order by TimeGenerated descSPL
sourcetype="openattacks:audit"
EventType IN ("data.export", "agent.external.send")
| table _time EventType Actor Asset Action Target CorrelationId
| sort - _timeInvestigation steps
- Join the resource read and outbound action by session, identity or process, preserving a bounded time window.
- Inspect destination ownership, payload classification and the user's intended action; baseline sanctioned sharing.
Technical detailsTooling, impact
Attack context
- AI's role
- Agent / MCP as target
Tooling as reported
AI products, models and artifacts
- MCP servers/clients
Detection rules
- Review attack-specific indicatorsGUIDANCE
Monitor for the techniques described in MCP line jumping: attack before first tool use. Refer to sources for specific IOCs and TTPs.
Source defensive notes
TOFU pinning (mcp-context-protector); sanitize ANSI; never auto-trust tool metadata
Sources & research credit
2 sourcesResearch by Trail of Bits · Trail of Bits
Record & downloads
- Source record name
- MCP line jumping: attack before first tool use
- Entry ID
- trailofbits-mcp-line-jumping-2025-04
- Catalog review
- October 20, 2026
- Category
- attacks-on-ai
- AI involvement confidence
- confirmed