A previous account of the Google Ads MCP adapter described it as read-only. Version 0.0.6 introduced tools that can create or mutate campaign-related configuration and exposed them in the ordinary default tool set. A user or approval policy that had previously trusted the server identity for querying reports might discover that the same named integration…
Inspect the ClaimsCase · DiggingBeagle record
Google Ads MCP 0.0.6 exposed mutation tools by default despite a read-only description
The release added four Google Ads mutation tools to the default MCP tool listing, creating a policy mismatch for clients or reviewers who treated the adapter as strictly read-only.
Public disclosure: Nov 2, 2026
Google Ads MCP release 0.0.6 when clients automatically expose or authorize the newly introduced write operations; requires a credential with corresponding Google Ads API permissions.
Assessment profile
Assessment method- Exploitation
- not established
- Evidence
- Grade B
- Remediation
- not applicable
- AI role
- AGAINST AI
Basis and provenance
Google Ads release and merged pull request establish a product change, not a verified security incident. Mutation endpoints are default-listed, but the proposal includes validateOnly first and an explicit confirmation stage. Existing Google Ads OAuth authorization still applies. No hostile operation, bypass or victim was observed.
Assessed Nov 11, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-R236-GOOGLEADS-MCP-DEFAULT-WRITE-TOOLS-01 · CLM-R236-GOOGLEADS-MCP-DEFAULT-WRITE-TOOLS-02 · CLM-ENR237-GOOGLEADS-CONFIRM-FLOW · Google Ads MCP release 0.0.6: campaign write tools
At a glance
A benign feature release can enlarge an agent's functional authority even if no prompt injection occurs. The meaningful safety question is who must reapprove a changed executable tool contract.
Read the implicationsThe release alone cannot establish unauthorized user effects, privilege escalation or a vulnerability in Google Ads. Tool exposure and successful API mutation have separate preconditions. Whether a specific consuming agent host exposes the propose-and-confirm UI faithfully or incorrectly maps a prior broad server approval to the new mutation tools.
Limits and uncertaintyFull account
A previous account of the Google Ads MCP adapter described it as read-only. Version 0.0.6 introduced tools that can create or mutate campaign-related configuration and exposed them in the ordinary default tool set. A user or approval policy that had previously trusted the server identity for querying reports might discover that the same named integration now advertises write-capable operations. That is a change in the available tool surface; it is not proof a model secretly gained privileges at the Google Ads API level.
The actual Google Ads OAuth credential must already permit the requested mutation, and real authorization checks continue to apply. The upstream release documents the added tools, but not a measured malicious edit, user bypass or compromise. A defensible finding concerns persistent client trust and the mismatch between a read-only safety assumption and subsequently expanded tool descriptors.
Google Ads MCP release 0.0.6 when clients automatically expose or authorize the newly introduced write operations; requires a credential with corresponding Google Ads API permissions.
The release alone cannot establish unauthorized user effects, privilege escalation or a vulnerability in Google Ads. Tool exposure and successful API mutation have separate preconditions.
The November 2 release is a vendor feature change, not an observed attack. Pull request #139 added creation of advertising assets, video uploading and label management, with mutation tools enabled by default in tools_config.yaml. It also specifies a validateOnly=true proposal before an explicit user confirmation of each actual mutation. That second stage is central to the security interpretation: exposure of write tools does not by itself show that the tool made an unapproved advertising change.
Risk depends on what an agent host does with refreshed tool declarations. A client that treats the MCP server's old read-only reputation as a permanent approval could present those tools without adequately reviewing the new mutation types; the Google Ads API still requires a token with write authority. Operator testing should verify that the proposed account, target object and change preview match the final executed API request, and that denial or an expired approval cannot be bypassed through a replay.
Mechanism and trust boundary
- 01
Upgrade a previously trusted MCP integration
An operator updates the server to a release containing additional write tools.
Boundary: package update / MCP tool catalog
- 02
Reuse previous client approval assumptions
Agent discovers the new tools without an independent policy review.
Boundary: server identity / effect-level approval
- 03
Potentially issue authorized Ads mutations
Actual changes still require Google Ads account and OAuth scopes that permit writes.
Boundary: tool call / advertiser account
Claims & evidence
CLM-ENR237-GOOGLEADS-CONFIRM-FLOWGoogle Ads MCP v0.0.6 included default-enabled mutation tools, but its referenced implementation first sends validateOnly=true and calls for explicit user confirmation before the actual mutate request.supported
Basis: reported finding
- supportsGoogle Ads MCP release 0.0.6: campaign write toolsprimary disclosure
v0.0.6 release note referencing PR #139; PR #139 description: propose-and-confirm and tools_config.yaml
CLM-R236-GOOGLEADS-MCP-DEFAULT-WRITE-TOOLS-01The v0.0.6 release documents newly available Google Ads mutation tools in the standard MCP server tool set.supported
Basis: reported finding
- supportsGoogle Ads MCP release 0.0.6: campaign write toolsprimary disclosure
google-ads-mcp v0.0.6: Release Notes
CLM-R236-GOOGLEADS-MCP-DEFAULT-WRITE-TOOLS-02New tool declarations do not independently override Google Ads OAuth permissions or demonstrate a malicious edit.supported
Basis: reported finding
- supportsGoogle Ads MCP release 0.0.6: campaign write toolsprimary disclosure
v0.0.6: Tool definitions and release limitations
Implications
A benign feature release can enlarge an agent's functional authority even if no prompt injection occurs. The meaningful safety question is who must reapprove a changed executable tool contract.
Controls and mitigations
- Pin adapter versions and inspect added or modified MCP tools before a trusted server is updated.
- Require approval by exact operation and specific advertising account for mutations, rather than blanket trust in the server name.
- Use separate API credentials for analytics and campaign management, with the read-only integration unable to perform writes even after tool-surface changes.
Unknowns and contradictions
- The release alone cannot establish unauthorized user effects, privilege escalation or a vulnerability in Google Ads. Tool exposure and successful API mutation have separate preconditions.
- Whether a specific consuming agent host exposes the propose-and-confirm UI faithfully or incorrectly maps a prior broad server approval to the new mutation tools.
Sources and citation
Material revision history
- Nov 11, 2026 · Published version · first publication · revision 238