server-memory
npm:@modelcontextprotocol/server-memory@2026.1.26
github.com/modelcontextprotocol/servers
Severity breakdown
Worst finding
Tool `create_entities` name implies a side effect that is not declared
· create_entities
`create_entities` looks like a side-effecting tool (its name contains a mutation verb), but its `side_effects` declaration is []. A policy synthesizer cannot produce safe rules for this tool because it cannot tell what it actually does.
fix: Declare the tool's true side effects explicitly. If the tool is genuinely read-only, rename it to match (e.g. `email.preview` rather than `email.send`).
All 6 findings
- highTool `create_entities` name implies a side effect that is not declared· create_entitiesexcessive agency
`create_entities` looks like a side-effecting tool (its name contains a mutation verb), but its `side_effects` declaration is []. A policy synthesizer cannot produce safe rules for this tool because it cannot tell what it actually does.
fix: Declare the tool's true side effects explicitly. If the tool is genuinely read-only, rename it to match (e.g. `email.preview` rather than `email.send`).
- highTool `create_relations` name implies a side effect that is not declared· create_relationsexcessive agency
`create_relations` looks like a side-effecting tool (its name contains a mutation verb), but its `side_effects` declaration is []. A policy synthesizer cannot produce safe rules for this tool because it cannot tell what it actually does.
fix: Declare the tool's true side effects explicitly. If the tool is genuinely read-only, rename it to match (e.g. `email.preview` rather than `email.send`).
- highTool `delete_entities` name implies a side effect that is not declared· delete_entitiesexcessive agency
`delete_entities` looks like a side-effecting tool (its name contains a mutation verb), but its `side_effects` declaration is []. A policy synthesizer cannot produce safe rules for this tool because it cannot tell what it actually does.
fix: Declare the tool's true side effects explicitly. If the tool is genuinely read-only, rename it to match (e.g. `email.preview` rather than `email.send`).
- highTool `delete_observations` name implies a side effect that is not declared· delete_observationsexcessive agency
`delete_observations` looks like a side-effecting tool (its name contains a mutation verb), but its `side_effects` declaration is []. A policy synthesizer cannot produce safe rules for this tool because it cannot tell what it actually does.
fix: Declare the tool's true side effects explicitly. If the tool is genuinely read-only, rename it to match (e.g. `email.preview` rather than `email.send`).
- highTool `delete_relations` name implies a side effect that is not declared· delete_relationsexcessive agency
`delete_relations` looks like a side-effecting tool (its name contains a mutation verb), but its `side_effects` declaration is []. A policy synthesizer cannot produce safe rules for this tool because it cannot tell what it actually does.
fix: Declare the tool's true side effects explicitly. If the tool is genuinely read-only, rename it to match (e.g. `email.preview` rather than `email.send`).
- mediumTool `search_nodes` accepts unconstrained string input· search_nodesunconstrained input
The following string parameter(s) have no `maxLength` constraint: `query`. Unbounded strings let an attacker stuff arbitrary payloads through the tool, including indirect-injection content.
fix: Add a `maxLength` to each string property, or constrain with an `enum` or `pattern`. Most legitimate tool inputs fit under a few hundred bytes.
Reproduce this
Every finding above comes from a public, deterministic rule engine — no LLM in the decision path, same input always the same output. You can run it against this server yourself and get byte-identical results:
curl -fsSL capframe.ai/install | sh capframe find ./your-mcp-config.json --out findings.json
Five-step quickstart if you want the whole Find → Bind → Guard loop. If you'd rather have someone read your real tool definitions rather than the advertised surface — including the ones this public scan can't reach — that's the Agent Security Audit.
How this was scored
Source sandbox — live tools/list captured in an ephemeral Docker container (parameter schemas included → R1/R2/R4 fire). Findings are emitted by the public capframe.findings.v1 schema. Score = 100 − (10·Critical + 4·High + 2·Medium + 1·Low), clamped to [0, 100].
Disagree with a finding? Open an issue.