ClawMem
On-device memory layer for AI agents. Claude Code, OpenClaw and Hermes. Hooks + MCP server + hybrid RAG search.
Caution. Specific findings reduced this grade. They are listed on the page. Grades reflect the full trust pyramid: code, verification depth, and reputation. New projects cap at C until adoption is earned.
How we verified
Verified is a snapshot. Live keeps it current, and builds your track record.
⚡ Connect GitHub → continuous verification on every pushwhy connect →Who stands behind it
yoloshii
Source: github_topic · also listed on Glama
Claim it to get a verified publisher badge, a free copy of our full audit findings, and direct contact for any high-priority issues we find. Or connect your repo for our deepest verification, Live Monitored: read-only, revoke anytime. What we access →
Install from
The grade above is for the source repository. Registries can serve a different version, so we mark the ones we were not able to read.
Disclosed vulnerabilities in this server's declared npm dependencies (via OSV). Whether each is reachable depends on the installed versions.
@modelcontextprotocol/sdk has cross-client data leak via shared server/transport instance reuse
Anthropic's MCP TypeScript SDK has a ReDoS vulnerability
yaml is vulnerable to Stack Overflow via deeply nested YAML collections
CLAUDE_SESSION_IDCLAWMEM_API_TOKENsecret ./bin/clawmem serve # with bearer token authCLAWMEM_CONFIG_DIRCLAWMEM_CONTRADICTION_POLICYlink v0.7.1. Merge-time contradiction gate policy. link inserts a new row + contradicts edge (default). supersede marks the old row status='inactive'.CLAWMEM_EMBED_API_KEYjina_your-key-hereCLAWMEM_EMBED_DIMENSIONSOpenAI https://api.openai.com text-embedding-3-small 1536 8K context, Matryoshka dimensions viaCLAWMEM_EMBED_MAX_CHARSNote: Cloud providers handle their own context window limits — ClawMem skips client-side truncation when an API key is set. Local llama-server truncates at (default: 6000 chars).CLAWMEM_EMBED_MODELjina-embeddings-v5-text-smallCLAWMEM_EMBED_TPM_LIMIT100000 Tokens-per-minute limit for cloud embedding pacing. Match to your provider tier.CLAWMEM_EMBED_URLexport =http://gpu-host:8088CLAWMEM_ENABLE_AMEMenabled A-MEM note construction + link generation during indexingCLAWMEM_FOCUS_ROOTCLAWMEM_HOOK_DEDUP_WINDOW_SECCLAWMEM_LLM_MODELexport =qwen3CLAWMEM_LLM_NO_THINKtrue Append /no_think to remote LLM prompts. Set to false for standard OpenAI models and other endpoints that reject or treat the Qwen-style suffix as literal prompt text.CLAWMEM_LLM_REASONING_EFFORTCLAWMEM_LLM_URLexport =http://gpu-host:8089CLAWMEM_MERGE_GUARD_DRY_RUNfalse v0.7.1. When true, Phase 2 merge-safety rejections are logged but not enforced — use for calibration before enabling the gate.CLAWMEM_MERGE_SCORE_NORMAL0.93 v0.7.1. Phase 2 consolidation merge-safety threshold when candidate and existing anchors align. Merges above this normalized 3-gram cosine score are allowed.CLAWMEM_NO_LOCAL_MODELSTo prevent fallback and fail fast instead, set =true.CLAWMEM_NUDGE_INTERVALCLAWMEM_PRECOMPACT_PROXIMITY_RATIOCLAWMEM_PROFILEbalanced # speed balanced deepCLAWMEM_SESSION_FOCUSCLAWMEM_SESSION_IDCLAWMEM_STDIO_MODECLAWMEM_VAULTSexport ='{"work":"~/.cache/clawmem/work.sqlite","personal":"~/.cache/clawmem/personal.sqlite"}'INDEX_PATHNO_COLORTool annotations
No tools have read-only/destructive annotations
Add readOnlyHint or destructiveHint annotations to every tool so hosts can warn users before invoking.
All four hints declared on every tool
31/31 tools missing one or more hints — __IMPORTANT (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint); memory_retrieve (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint); search (missing: readOnlyHint, destructiveHint, idempotentHint, openWorldHint), +28 more. OpenAI's directory rejects tools where any of the four hints are missing or non-boolean.
For every tool, set all four hints (readOnlyHint, destructiveHint, idempotentHint, openWorldHint) to explicit true/false values that match the handler’s actual behaviour.
Tool inputs are validated
29/31 tool handlers declare input schemas (94%)
Declare an inputSchema with zod/joi/yup on every tool definition.
Tool handlers catch errors
Only 4/31 tool handlers wrap calls in try/catch (13%)
Wrap each tool handler body in try/catch and return a structured error response.
Tool test coverage
Only 6/31 tools referenced in tests (19%)
Write tests that reference each tool by name so every tool has at least one test.
Shell command execution
150 child_process calls — runs shell commands
Prefer library functions over shell-outs. If you must shell out, ensure all inputs are properly escaped.
Secrets not logged
1 secret value sent to console.log
Redact or omit secret values from log output.
Production dependencies are patched
0 critical, 2 high severity in production deps — @modelcontextprotocol/sdk@1.25.1 (high), @modelcontextprotocol/sdk@1.25.1 (high)
Run npm audit fix, or upgrade the affected packages to a non-vulnerable version.
Dependency freshness
1/6 production deps stale: gray-matter@2023-07-12 (2.9y)
Claim the listing to review these findings one by one and send us a correction where you disagree, straight to the team. Claiming also means we tell you when the grade moves, and reach you first if we find anything urgent.
[](https://m8ven.ai/mcp/yoloshii/clawmem)?variant=verified from the URL.Vetting this one by hand? Tool Check is an MCP that scores other MCPs. Add it once and ask Claude, ChatGPT, or any MCP client to grade a server, surface CVEs, check the publisher, and suggest safer alternatives — before you install.
https://m8ven.ai/api/mcp/tool-check