mcp-behave@0.4.5
CAUTION· not startedPublisher claims: "Runtime behavioral auditor for MCP servers — scope-violation detection via native libc interposition & runtime tracing" Observed in sandbox: no network, touches nothing outside the project, writes no memory.
Coverage
Not started: dependency install failed; the server itself was not observed.- Tested
- the dependency install step · every file read, file written and host contacted by what ran
- Not tested
- server startup · the tool list · tool behavior
What this means for the label: Caution here covers what ran. The server's tools were not exercised with a real credential, so nothing about their behavior is asserted; the credential summary below says what to do with that.
Before you give it your token
- It will ask for
MCP_BEHAVE_TOKENConfiguration, not credentials: MCP_BEHAVE_ALLOW_ALL_EXEC, MCP_BEHAVE_ALLOWED_COMMANDS, MCP_BEHAVE_CALL_TIMEOUT, MCP_BEHAVE_DASHBOARD_URL, MCP_BEHAVE_DEBUG, MCP_BEHAVE_EBPF_REQUIRED, MCP_BEHAVE_MOCK_EBPF, MCP_BEHAVE_STATE_DIR, MCP_BEHAVE_UNSAFE_HOST, OUT_DIR, PROBE_NOISE_EXTRA, TRACE_FILE, XDG_STATE_HOME- Where it talks to
- mcp-behave.dev (its own)Also githubusercontent.com, example.org, example.net, modelcontextprotocol-security.io — not the publisher's domain; check what for before connecting.
- What we saw when it ran
- no credential files read · no network
- What we could not test
- server startup · the tool list · tool behavior
Connect it with a scoped credential, under the guard, and read the third-party hosts first. The code names githubusercontent.com, example.org, example.net, modelcontextprotocol-security.io as well as its own vendor. The guard injects MCP_BEHAVE_TOKEN only into requests to mcp-behave.dev (and its subdomains) and refuses the rest.
Never paste a credential into a chat, a ticket or a report, including ours. How the guard holds it.
What changed since 0.3.0
vet-skills diff mcp-behave 0.3.0 0.4.5This version reaches for something 0.3.0 did not: 3 new hosts (example.net, example.org, mcp-behave.dev) · 2 new credential files (~/.ssh/config, ~/.ssh/known_hosts). Compared from the two recorded runs, not from source.
- New hosts
- example.net, example.org, mcp-behave.dev
- New credential files
- ~/.ssh/config, ~/.ssh/known_hosts
- No longer contacted
- astral.sh, attacker-c2.example, python.org
0.4.5 was not run in the sandbox (install failed); the comparison covers what static analysis recorded only.
The guard can hold an update like this until someone approves it (runtime.on_change: hold). A change that only shows with a real credential, or on a path the run did not reach, is not in a trace and so not in this comparison.
Is [email protected] safe to install?
KillerSkills ran it in the KillerSkills honeypot sandbox (2 scenarios; not started: dependency install failed; the server itself was not observed) on 2026-11-20 and scored it 72/100 (Caution). It is not known to be malicious, but it has behaviour or code a security policy should decide on before install. Notable findings: Access to sensitive file paths; Access to sensitive file paths; Access to sensitive file paths. Summary: no network, touches nothing outside the project, writes no memory.
Behavior trace
Honeypot run · 1 s · 2 scenarios · 0 deviationsStatic findings
return True, "/etc/shadow"Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.if ".netrc" in parts:Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.".env", "id_rsa", "id_ed25519", ".ssh", ".pem", ".key", "private_key",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.if norm == "/etc/shadow" or norm.endswith("/etc/shadow"):Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.f.write("[default]\naws_access_key_id = AKIAIOSFODNN7EXAMPLE\naws_secret_access_key = wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY\n")Fix: Never read or enumerate credential locations; accept a documented environment variable instead.(.ssh/id_rsa, .aws/credentials, .env, .netrc) so credential probes trip detectionFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.return True, ".netrc"Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.elif any(sens in arg_norm for sens in ("/.ssh", "/.aws", "/.kube", "/.gnupg", "/etc/shadow")):Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.".aws", ".kube", "/etc/passwd", "/etc/shadow", ".npmrc", ".git/config",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.the server. This runs as YOU -- no sudo needed. Two things can prevent it:Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.subprocess.Popen(["osascript", "-e", script], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.[!] HIGH prompt injection pattern detected in tool 'notes_analyzer' description: 'Ignore previous instructions'Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.SENSITIVE = (".ssh", "id_rsa", "id_ed25519", ".env", ".aws", "credentials",Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.subprocess.Popen(["osascript", "-e", script], stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.with open(os.path.join(canary_dir, ".env"), "w") as f:Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.res = subprocess.run(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.# 1. NPX / Node.jsFix: Pin versions and sources; vendor anything fetched at install time.declared_paths.add(os.path.expanduser("~/.ssh/config"))Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.with open(os.path.join(canary_dir, ".netrc"), "w") as f:Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.("../../../../../../../../etc/passwd", -1),Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.".netrc", "/etc/shadow", ".kube", ".docker/config")Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose."`python -m mcp_server_fetch`). For --transport sse/"Fix: Namespace tool names and keep tool descriptions literal; no instructions to the agent inside them.- Unpinned remote package execution (e.g. npx without version tag).Fix: Pin versions and sources; vendor anything fetched at install time.- Direct prompt injection patterns (e.g., `Ignore previous instructions`, `SYSTEM PROMPT:`, `<system>`, `[system instruction]`, `you must always call this tool first`).Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.subprocess.run([cc, "-shared", "-fPIC", "-O2", "-o", so, src, "-ldl"],Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.out = subprocess.run([cs, "-d", "--verbose=4", path],Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.subprocess.run([cc, *arch_flags, "-dynamiclib", "-O2",Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.res = subprocess.run(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out."""Inspect local NPX runner arguments and local disk cache offline (or query registry if allow_network=True)."""Fix: Pin versions and sources; vendor anything fetched at install time.declared_paths.add(os.path.expanduser("~/.ssh/known_hosts"))Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.- Direct prompt injection patterns (e.g., `Ignore previous instructions`, `SYSTEM PROMPT:`, `<system>`, `[system instruction]`, `you must always call this tool first`).Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.help="Path to manifest.json, tools.json, output directory, or '-' for stdin (default: stdin).",Fix: Remove the call, or declare the host in frontmatter and explain in the README what is sent and why.Cost to the agent
Tokens estimated at four characters each. The tool list is re-read by the agent on every turn, so it is the number to compare between servers that do the same job: ranked across the index.
Known vulnerabilities
Not checked against OSV yet; the weekly check has not reached this version.
What would make this Verified
8 points to 80 · weights: behavior 45%, static 25%, prompt 15%, publisher 15%- Unclaimed publisher≈ +8 on the totalClaim navid72m: a verified publisher gets +50 on publisher trust. It is free and usually instant with GitHub sign-in.
- Access to sensitive file paths≈ +4 on the totalStop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.
- AWS access key pattern detected≈ +4 on the totalNever read or enumerate credential locations; accept a documented environment variable instead.
- Privileged sudo command≈ +4 on the totalBuild commands from fixed arguments; do not interpolate untrusted strings into a shell.
- Python subprocess execution detected≈ +2 on the totalBuild commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
- npx invocation of non-well-known package≈ +2 on the totalPin versions and sources; vendor anything fetched at install time.