OWASP MCP Top 10 — Compliance Assessment

2025 · 2026-04-23 registry entry

Server
compute.googleapis.com (compute-googleapis-com-20260728163042-10277d)
Scan id
39b117c1-9a86-4b15-bc6d-f467555c73b6
Assessed at
2026-07-28 16:30:43 UTC
Sentinel version
0.4.0
DRAFT for review — not legal advice. See attestation block for verification instructions.

Executive summary

✗ Non-compliant

Assessment of compute.googleapis.com against OWASP MCP Top 10: overall status non compliant. Of 10 controls, 0 met, 1 unmet, 1 partial, 8 not assessed, 0 not applicable. 2 control(s) had complete mapped-assessor execution evidence; 8 lacked sufficient execution evidence and 0 were outside the current assessor mapping. Unmet controls have findings at or above the framework's mandatory severity threshold and should be remediated before relying on this server in a regulated deployment. All claims are traceable to individual finding rows via finding_id and to the governing rule via rule_id; the enclosing signed envelope commits MCP Sentinel to the exact bytes of this report.

Coverage & transparency

Coverage band
low
Coverage ratio
32%
Rules version
2026-04-23

Analysis techniques applied:

  • capability-graph
  • composite
  • linguistic
  • schema-inference
  • similarity
  • structural
  • stub
  • unicode

Controls (10)

IDControlStatusEvidence
MCP01 Prompt Injection ? Not assessed 0
MCP02 Tool Poisoning ✗ Unmet 1
MCP03 Command Injection ? Not assessed 0
MCP04 Data Exfiltration ? Not assessed 0
MCP05 Privilege Escalation ? Not assessed 0
MCP06 Excessive Permissions ? Not assessed 0
MCP07 Insecure Configuration ! Partial 30
MCP08 Dependency Vulnerabilities ? Not assessed 0
MCP09 Logging & Monitoring Failures ? Not assessed 0
MCP10 Supply Chain Compromise ? Not assessed 0

Control details

MCP01 — Prompt Injection

? Not assessed

0/24 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: A1 (execution-ledger); A5 (execution-ledger); A7 (execution-ledger); A8 (execution-ledger); A9 (execution-ledger); B5 (execution-ledger); F1 (execution-ledger); F6 (execution-ledger); G1 (execution-ledger); G2 (execution-ledger); G3 (execution-ledger); G4 (execution-ledger); G5 (execution-ledger); H2 (execution-ledger); H3 (execution-ledger); I3 (execution-ledger); I6 (execution-ledger); I7 (execution-ledger); J3 (execution-ledger); J5 (execution-ledger); J6 (execution-ledger); N4 (execution-ledger); N9 (execution-ledger); N12 (execution-ledger).

Framework reference

Control text:

Attacker-controlled content delivered through MCP server descriptions, parameters, resources, prompts, or initialize metadata that hijacks agent reasoning. Includes direct and indirect injection.

MCP02 — Tool Poisoning

✗ Unmet

0/15 mapped assessor rule(s) are proven to have executed; 1 finding(s) observed (1 critical); at least one finding is at or above the high threshold (status: unmet). 15 mapped assessor(s) did not execute, so this finding-backed result does not imply complete control coverage.

Framework reference

Control text:

Tools that misrepresent their behaviour through deceptive names, annotations, namespace squatting, homoglyphs, or time-based drift (rug pulls).

  • Critical Tool Annotation Deception (finding f9eb7e57-7328-476c-a971-e30f5b5f9faa, confidence 85%) SOURCE: external-content at tool list_instance_attached_disks — The MCP spec (2025-03-26) treats tool annotations as server-declared hints that AI clients may use to skip user confirmation. readOnlyHi
Required mitigations
  • Remove readOnlyHint: true from any tool whose parameters or description indicate write, delete, or overwrite capability. If the tool is genuinely read-only, rename the misleading parameter or update the description. If the tool is destructive, set destructiveHint: true so AI clients require user confirmation before invocation. AI clients (ChatGPT, Cursor, Roo Code, JetBrains Copilot per Invariant Labs 2025) auto-approve tools declaring readOnlyHint: true without cross-checking the schema — a deceptive annotation bypasses the strongest consent control in the MCP protocol.

MCP03 — Command Injection

? Not assessed

0/6 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: C1 (execution-ledger); C9 (execution-ledger); C13 (execution-ledger); C16 (execution-ledger); J2 (execution-ledger); J7 (execution-ledger).

Framework reference

Control text:

Server code that executes shell commands, system calls, or git operations built from attacker-controlled input without sanitisation.

MCP04 — Data Exfiltration

? Not assessed

0/15 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: A3 (execution-ledger); F1 (execution-ledger); F3 (execution-ledger); F7 (execution-ledger); G7 (execution-ledger); I9 (execution-ledger); I13 (execution-ledger); K18 (execution-ledger); O4 (execution-ledger); O5 (execution-ledger); O6 (execution-ledger); O8 (execution-ledger); O9 (execution-ledger); O10 (execution-ledger); P3 (execution-ledger).

Framework reference

Control text:

Server features that permit sensitive data to leave the trust boundary via HTTP, DNS, headers, URIs, or covert channels.

MCP05 — Privilege Escalation

? Not assessed

0/7 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: C2 (execution-ledger); C8 (execution-ledger); C10 (execution-ledger); C12 (execution-ledger); I4 (execution-ledger); I12 (execution-ledger); J1 (execution-ledger).

Framework reference

Control text:

Vertical privilege increases via JWT confusion, deserialization, evaluation of untrusted input, OAuth scope widening, or capability escalation post-initialization.

MCP06 — Excessive Permissions

? Not assessed

0/7 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: A2 (execution-ledger); B3 (execution-ledger); B7 (execution-ledger); E4 (execution-ledger); F2 (execution-ledger); I11 (execution-ledger); I16 (execution-ledger).

Framework reference

Control text:

Tools or roots claiming broader filesystem, network, or credential access than required for their stated function.

MCP07 — Insecure Configuration

! Partial

0/11 mapped assessor rule(s) are proven to have executed; 30 finding(s) observed (30 medium); all findings are below the high threshold (status: partial). 11 mapped assessor(s) did not execute, so this finding-backed result does not imply complete control coverage.

Framework reference

Control text:

Servers shipped with insecure defaults: no authentication, wildcard CORS, HTTP transport, weak session tokens, disabled TLS validation, empty sandboxes.

  • Medium Schema Allows Unconstrained Additional Properties (finding e5dd1b89-57b6-41b8-890b-b3f08a5c9744, confidence 75%) SOURCE: user-parameter at tool create_instance — Tool "create_instance" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that
  • Medium Schema Allows Unconstrained Additional Properties (finding cca23fe0-7b1f-4e3a-a238-f8f990ab2cda, confidence 75%) SOURCE: user-parameter at tool delete_instance — Tool "delete_instance" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that
  • Medium Schema Allows Unconstrained Additional Properties (finding 4713c590-66a3-4dfc-9532-c6b7559fd67a, confidence 75%) SOURCE: user-parameter at tool start_instance — Tool "start_instance" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that by
  • Medium Schema Allows Unconstrained Additional Properties (finding 8bc1f684-a322-408c-845f-81e109781d8d, confidence 75%) SOURCE: user-parameter at tool stop_instance — Tool "stop_instance" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that bypa
  • Medium Schema Allows Unconstrained Additional Properties (finding 6f40d60c-b25b-4bb4-a6aa-9bd8dd56b1ef, confidence 75%) SOURCE: user-parameter at tool reset_instance — Tool "reset_instance" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that by
  • Medium Schema Allows Unconstrained Additional Properties (finding a9ca73fd-9375-439d-92db-2e7d784186b2, confidence 75%) SOURCE: user-parameter at tool get_instance_basic_info — Tool "get_instance_basic_info" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undec
  • Medium Schema Allows Unconstrained Additional Properties (finding b10266fa-5957-4c41-9d8d-74c0f7472043, confidence 75%) SOURCE: user-parameter at tool set_instance_machine_type — Tool "set_instance_machine_type" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read u
  • Medium Schema Allows Unconstrained Additional Properties (finding 51b4f92c-3a9d-4a6f-a89c-11a22432e857, confidence 75%) SOURCE: user-parameter at tool list_instance_attached_disks — Tool "list_instance_attached_disks" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may
  • Medium Schema Allows Unconstrained Additional Properties (finding 98a4089c-f97c-4e08-a53e-309184fcb8d6, confidence 75%) SOURCE: user-parameter at tool list_instances — Tool "list_instances" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that by
  • Medium Schema Allows Unconstrained Additional Properties (finding 592ee716-fdcc-4019-bac6-ca6335c74ec0, confidence 75%) SOURCE: user-parameter at tool get_instance_group_manager_basic_info — Tool "get_instance_group_manager_basic_info" input_schema accepts arbitrary extra keys. The declared properties are validated, bu
  • Medium Schema Allows Unconstrained Additional Properties (finding 117ff4d0-a119-4f6b-8c83-bf7d7cf68a53, confidence 75%) SOURCE: user-parameter at tool list_instance_group_managers — Tool "list_instance_group_managers" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may
  • Medium Schema Allows Unconstrained Additional Properties (finding a9eb034b-403e-47a3-a05a-e3ade9c9c9e8, confidence 75%) SOURCE: user-parameter at tool list_managed_instances — Tool "list_managed_instances" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undecla
  • Medium Schema Allows Unconstrained Additional Properties (finding ba57e051-28c3-478f-a3b2-2762b822c9e5, confidence 75%) SOURCE: user-parameter at tool list_instance_templates — Tool "list_instance_templates" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undec
  • Medium Schema Allows Unconstrained Additional Properties (finding b12906bf-c95e-4b92-b8f5-57b21c30c9a2, confidence 75%) SOURCE: user-parameter at tool get_instance_template_basic_info — Tool "get_instance_template_basic_info" input_schema accepts arbitrary extra keys. The declared properties are validated, but the hand
  • Medium Schema Allows Unconstrained Additional Properties (finding 34e25997-f84d-4124-991e-7233265db93a, confidence 75%) SOURCE: user-parameter at tool get_instance_template_properties — Tool "get_instance_template_properties" input_schema accepts arbitrary extra keys. The declared properties are validated, but the hand
  • Medium Schema Allows Unconstrained Additional Properties (finding 22cd2e9a-e75e-45a5-a6a4-4adef5316948, confidence 75%) SOURCE: user-parameter at tool get_disk_basic_info — Tool "get_disk_basic_info" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared ke
  • Medium Schema Allows Unconstrained Additional Properties (finding 444b65a1-2a28-4916-b55b-b7084744ad02, confidence 75%) SOURCE: user-parameter at tool get_disk_performance_config — Tool "get_disk_performance_config" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may re
  • Medium Schema Allows Unconstrained Additional Properties (finding 65d2b5d7-9a77-407e-bd6f-5d20e51617d2, confidence 75%) SOURCE: user-parameter at tool list_disks — Tool "list_disks" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that bypass eve
  • Medium Schema Allows Unconstrained Additional Properties (finding 0769aee4-219b-4d9c-bf4d-99ea8db55d57, confidence 75%) SOURCE: user-parameter at tool list_accelerator_types — Tool "list_accelerator_types" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undecla
  • Medium Schema Allows Unconstrained Additional Properties (finding 3be2bbe4-e85a-4fac-a4d0-bd34e989d7c7, confidence 75%) SOURCE: user-parameter at tool list_machine_types — Tool "list_machine_types" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys
  • Medium Schema Allows Unconstrained Additional Properties (finding 2c65db68-2026-49f1-a6a7-c58281d6d6dd, confidence 75%) SOURCE: user-parameter at tool list_images — Tool "list_images" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that bypass e
  • Medium Schema Allows Unconstrained Additional Properties (finding 52c52d69-5d6d-40c5-8465-964fc467aaa8, confidence 75%) SOURCE: user-parameter at tool get_zone_operation — Tool "get_zone_operation" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys
  • Medium Schema Allows Unconstrained Additional Properties (finding 1ae01d30-5637-45c0-8a47-a7238d99e9fc, confidence 75%) SOURCE: user-parameter at tool get_reservation_basic_info — Tool "get_reservation_basic_info" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read
  • Medium Schema Allows Unconstrained Additional Properties (finding d5e9ab90-d6f3-42cd-acce-194b33a117fc, confidence 75%) SOURCE: user-parameter at tool get_reservation_details — Tool "get_reservation_details" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undec
  • Medium Schema Allows Unconstrained Additional Properties (finding fe5ac821-00f7-4f03-b993-8e01d8662b57, confidence 75%) SOURCE: user-parameter at tool list_reservations — Tool "list_reservations" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys t
  • Medium Schema Allows Unconstrained Additional Properties (finding 2db588c0-713f-4c07-a305-1b91ac7d3370, confidence 75%) SOURCE: user-parameter at tool list_commitments — Tool "list_commitments" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys tha
  • Medium Schema Allows Unconstrained Additional Properties (finding 762d401f-48c7-4b82-8660-331775c01491, confidence 75%) SOURCE: user-parameter at tool get_commitment_basic_info — Tool "get_commitment_basic_info" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read u
  • Medium Schema Allows Unconstrained Additional Properties (finding ebd36c86-f290-47ad-9170-57800c5476ca, confidence 75%) SOURCE: user-parameter at tool list_commitment_reservations — Tool "list_commitment_reservations" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may
  • Medium Schema Allows Unconstrained Additional Properties (finding 4c303000-030e-48ab-9d25-b7e59b0832a6, confidence 75%) SOURCE: user-parameter at tool list_snapshots — Tool "list_snapshots" input_schema accepts arbitrary extra keys. The declared properties are validated, but the handler may read undeclared keys that by
  • Medium No Authentication Required (finding 845971fb-be88-4217-9ce3-43fcf7983964, confidence 75%) SOURCE: environment at capability:tools — An MCP server that answers tool enumeration without authentication trusts the network. Under modern threat models (CCS 2007 DNS rebinding, open cloud networki
Required mitigations
  • Set additionalProperties: false on every object schema. This rejects any key outside the declared properties, closing the side-channel smuggling path and enforcing the schema's stated contract.
  • Require authentication for all MCP server connections. For remote MCP servers adopt OAuth 2.0 per RFC 9700 / the MCP Authorization specification. For stdio-launched servers rely on the parent process's security boundary and DO NOT expose the same server over network transports. Even localhost-bound servers should require auth: DNS rebinding (CCS 2007) makes localhost reachable from any browser tab.

MCP08 — Dependency Vulnerabilities

? Not assessed

0/7 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: D1 (execution-ledger); D2 (execution-ledger); D3 (execution-ledger); D4 (execution-ledger); D5 (execution-ledger); D6 (execution-ledger); D7 (execution-ledger).

Framework reference

Control text:

Known-CVE, abandoned, malicious, or typosquatted dependencies shipped with the server.

MCP09 — Logging & Monitoring Failures

? Not assessed

0/6 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: C6 (execution-ledger); E3 (execution-ledger); K1 (execution-ledger); K2 (execution-ledger); K3 (execution-ledger); K20 (execution-ledger).

Framework reference

Control text:

Absent, insufficient, or tamperable audit logging that prevents incident detection and response.

MCP10 — Supply Chain Compromise

? Not assessed

0/27 mapped assessor rule(s) are proven to have executed; no adverse findings were observed, but the control is not assessed because execution evidence is incomplete. Unexecuted assessors: A4 (execution-ledger); D3 (execution-ledger); D5 (execution-ledger); D7 (execution-ledger); F5 (execution-ledger); I5 (execution-ledger); J7 (execution-ledger); K9 (execution-ledger); K10 (execution-ledger); K11 (execution-ledger); L1 (execution-ledger); L2 (execution-ledger); L3 (execution-ledger); L4 (execution-ledger); L5 (execution-ledger); L6 (execution-ledger); L7 (execution-ledger); L8 (execution-ledger); L9 (execution-ledger); L10 (execution-ledger); L11 (execution-ledger); L12 (execution-ledger); L13 (execution-ledger); L14 (execution-ledger); L15 (execution-ledger); Q4 (execution-ledger); Q13 (execution-ledger).

Framework reference

Control text:

Compromise of the build, publish, or distribution pipeline upstream of the server — including CI/CD token theft, build-artifact tampering, and manifest confusion.

Multi-step attack chains

No multi-step attack chains were synthesized for this server.

Cryptographic attestation

Algorithm
HMAC-SHA256
Key ID
mcps-hkdf-e8990a4e1526
Signer
mcp-sentinel/v1
Signed at
2026-07-28T16:30:43.290Z
Canonicalization
RFC8785

HMAC-SHA256 signature (base64, wrapped at 64 chars):

Xs4Ru3BGA7RXfxRposcJZ9+O69c+7hRlJm0rOT/FeYs=

Verification instructions:

To verify this report:
  1. Extract the report body (everything except the .attestation field).
  2. Canonicalize the body via RFC 8785 (JCS).
  3. Compute HMAC-SHA256 with the signing key for key_id "mcps-hkdf-e8990a4e1526".
  4. Base64-encode the result and compare with the signature above.