converse-mcp-server@4.13.0
VERIFIEDPublisher claims: "Converse MCP Server - Converse with other LLMs with chat and consensus tools" Observed in sandbox: no network, touches nothing outside the project, writes no memory.
Coverage
Server started and answered tools/list.- Tested
- MCP initialize and tools/list · every tool description, for hidden instructions · every file read, file written and host contacted by what ran
- Not tested
- what the tools do when called with real credentials
Before you give it your token
- It will ask for
ANTHROPIC_API_KEYAPI_KEY_ENV_VARSCLAUDE_CODE_OAUTH_TOKENCLEF_CONTEXT_TOKENSMAX_MCP_OUTPUT_TOKENSConfiguration, not credentials: ASYNC_CACHE_DIR, ASYNC_DISK_TTL_MS, ASYNC_MEMORY_TTL_MS, CLAUDE_CONFIG_DIR, CLIENT_CWD, CODEX_HOME, COPILOT_CLI_PATH, INIT_CWD, MCP_TRANSPORT, npm_config_local_prefix, Path- Where it talks to
- no vendor host named in the codeAlso openrouter.ai, antigravity.google, typesafe.ai, cloudflare.com, abliteration.ai, deepseek.com, x.ai — 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
- what the tools do when called with real credentials
Connect it with a scoped credential, under the guard, and read the third-party hosts first. The code names openrouter.ai, antigravity.google, typesafe.ai, cloudflare.com, abliteration.ai, deepseek.com, x.ai as well as its own vendor. The guard injects ANTHROPIC_API_KEY, API_KEY_ENV_VARS, CLAUDE_CODE_OAUTH_TOKEN, CLEF_CONTEXT_TOKENS, MAX_MCP_OUTPUT_TOKENS only into requests to the host you bind (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 4.12.0
vet-skills diff converse-mcp-server 4.12.0 4.13.0In the sandbox, this version reached for nothing 4.12.0 did not. Compared from the two recorded runs, not from source.
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) on 2026-11-21 and scored it 81/100 (Verified). It did what it claims and nothing else we could detect: no undeclared network egress, no credential reads, no persistent instruction writes. Summary: no network, touches nothing outside the project, writes no memory.
Behavior trace
Honeypot run · 70 s · 2 scenarios · 0 deviationsStatic findings
// Convert messages to SDK input formatFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.// Load environment variables from appropriate .env fileFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// node builtins, so this stays a sub-100ms round trip for `npx converse -v`.Fix: Pin versions and sources; vendor anything fetched at install time.".env.example"Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.return existsSync(join(configDir, '.credentials.json'));Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// Fall back to .env for any missing variablesFix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// Priority: .env.test (for test env) > .env (default)Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.dotenv.config({ path: '.env.test', quiet: true });Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.// then schedules a post-kill grace timer so we never hang if onExit neverFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.prompt: '<s>[INST] Hello [/INST]',Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.// Extra wall-clock grace before the JS-side hard kill fires (ms).Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.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.