Skills/PyPI/research-mcp

research-mcp@0.1.0

CAUTION· not started

Publisher claims: "MCP server for editing Microsoft Word documents safely and citing from Zotero (Windows)" Observed in sandbox: no network, touches nothing outside the project, writes no memory.

Registry PyPI (MCP server)Publisher russell-day · publishing for 7 daysInstalls 151Last detonated 42 min ago

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
UNSLOTH_API_KEYZOTERO_API_KEY or --max-output-tokens / --new-tokenConfiguration, not credentials: RESEARCH_MCP_CONFIG, RESEARCH_MCP_HOME, RESEARCH_MCP_OWNED_WORD_FILE, UNSLOTH_BASE_URL, UNSLOTH_MODEL, ZOTERO_LIBRARY_ID
Where it talks to
no vendor host named in the codeAlso openxmlformats.org, microsoft.com, cloudflare.com, zotero.org, astral.sh, doi.org, crossref.org, unpaywall.org — not the publisher's domain; check what for before connecting.
What the honeypot saw
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 openxmlformats.org, microsoft.com, cloudflare.com, zotero.org, astral.sh, doi.org, crossref.org, unpaywall.org as well as its own vendor. The guard injects UNSLOTH_API_KEY, ZOTERO_API_KEY only into requests to the host you bind (and its subdomains) and refuses the rest.

$vet-skills run --net declared -- uvx research-mcp

Never paste a credential into a chat, a ticket or a report, including ours. How the guard holds it.

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 2026-11-15 and scored it 73/100 (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; Auto-update or dynamic fetch configuration; Access to sensitive file paths. Summary: no network, touches nothing outside the project, writes no memory.

Behavior trace

Honeypot run · 0 s · 2 scenarios · 0 deviations
t+CategoryObserved action
0.0sprocessno install scripts in bundle
0.1sprocessdependency install exit 1 ·
0.1sclaimednot started: dependency install failed (exit 1); the server itself was not observed
0.1sprocessscript backend runs setup only
0.1sprocessno install scripts to re-run

Static findings

MEDIUM
Process kill commandsrc/research_mcp/benchmark.py:722 · Process kill command — agent can terminate processes# append below used to hit UnboundLocalError and kill the wholeFix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.
AST05
HIGH
Access to sensitive file pathssrc/research_mcp/benchmark.py:929 · Access to sensitive file paths — SSH keys, AWS credentials, env files, or system filesdotenv = _read_dotenv(workspace / ".env")Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.
AST05
MEDIUM
Python subprocess execution detectedsrc/research_mcp/integration.py:306 · Python subprocess execution detectedhelper = subprocess.Popen(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Python subprocess execution detectedsrc/research_mcp/operation_process.py:46 · Python subprocess execution detectedcompleted = subprocess.run(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Python subprocess execution detectedsrc/research_mcp/benchmark.py:395 · Python subprocess execution detectedcompleted = subprocess.run(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
HIGH
Auto-update or dynamic fetch configurationsrc/research_mcp/remote.py:395 · Auto-update or dynamic fetch configuration — agent may load new code at runtimecommand = [str(exe), "tunnel", "--no-autoupdate", *config_args, "--url", f"http://127.0.0.1:{port}"]Fix: Pin versions and sources; vendor anything fetched at install time.
AST06
HIGH
Access to sensitive file pathssrc/research_mcp/benchmark.py:932 · Access to sensitive file paths — SSH keys, AWS credentials, env files, or system filesraise RuntimeError("UNSLOTH_API_KEY is missing from the environment and .env")Fix: Stop enumerating or reading credential locations; take secrets from a declared environment variable the user sets on purpose.
AST05
HIGH
Tunneling service URLsrc/research_mcp/remote.py:52 · Tunneling service URL — may route traffic to attacker infrastructure_TUNNEL_URL = re.compile(r"https://(?!api\.)[a-z0-9][a-z0-9-]*\.trycloudflare\.com")Fix: Replace encoded or assembled strings with plain literals and remove undeclared network calls.
AST07
MEDIUM
Python subprocess execution detectedsrc/research_mcp/recovery.py:148 · Python subprocess execution detectedsubprocess.Popen([str(ZOTERO_EXE)], close_fds=True)Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Python subprocess execution detectedtests/test_setup_cmd.py:726 · Python subprocess execution detectedresult = subprocess.run([sys.executable, str(script), "--version"], capture_output=True, text=True,Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Tunneling service URLtests/test_remote.py:38 · Tunneling service URL — may route traffic to attacker infrastructure2026-11-06T07:23:03Z INF | https://seven-random-words-here.trycloudflare.com |Fix: Replace encoded or assembled strings with plain literals and remove undeclared network calls.
AST07
MEDIUM
Tunneling service URLtests/test_remote.py:315 · Tunneling service URL — may route traffic to attacker infrastructureassert remote.parse_tunnel_url("see https://api.trycloudflare.com/tunnel") is NoneFix: Replace encoded or assembled strings with plain literals and remove undeclared network calls.
AST07
LOW
Prompt injectionREADME.md:64 · Prompt injection — known jailbreak technique reference · informational: not an instruction filein ChatGPT on the web with developer mode (per [OpenAI](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt),Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.
AST01
LOW
Authorization escalation languageREADME.md:67 · Authorization escalation language · informational: not an instruction file1. Settings > Apps (or Apps & Connectors) > Advanced settings: turn on **Developer mode**Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.
AST01
LOW
Authorization escalation languageREADME.md:64 · Authorization escalation language · informational: not an instruction filein ChatGPT on the web with developer mode (per [OpenAI](https://help.openai.com/en/articles/12584461-connecting-custom-mcp-servers-in-chatgpt),Fix: Remove the text that addresses the agent rather than the reader; describe behavior, do not instruct.
AST01
MEDIUM
Process kill commandsrc/research_mcp/operation_process.py:17 · Process kill command — agent can terminate processes# Never enumerate or kill arbitrary WINWORD processes.Fix: Build commands from fixed arguments; do not interpolate untrusted strings into a shell.
AST05
MEDIUM
Python subprocess execution detectedsrc/research_mcp/rendering.py:102 · Python subprocess execution detectedproc = subprocess.run(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Python subprocess execution detectedsrc/research_mcp/recovery.py:61 · Python subprocess execution detectedout = subprocess.run(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Tunneling service URLtests/test_remote.py:44 · Tunneling service URL — may route traffic to attacker infrastructureSAMPLE_HOST = "https://seven-random-words-here.trycloudflare.com"Fix: Replace encoded or assembled strings with plain literals and remove undeclared network calls.
AST07
MEDIUM
Python subprocess execution detectedsrc/research_mcp/word_service.py:106 · Python subprocess execution detectedsubprocess.Popen(Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Python subprocess execution detectedtests/test_remote.py:384 · Python subprocess execution detectedreturn subprocess.Popen([sys.executable, str(script), str(sample), mode], **kwargs)Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Python subprocess execution detectedsrc/research_mcp/recovery.py:135 · Python subprocess execution detectedsubprocess.run(["taskkill", "/PID", str(pid), "/F"], capture_output=True, text=True)Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
MEDIUM
Python subprocess execution detectedsrc/research_mcp/setup_cmd.py:384 · Python subprocess execution detectedreturn subprocess.run(command, capture_output=True, text=True, timeout=120)Fix: Build commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
AST05
LOW
Prompt injectionREADME.md:67 · Prompt injection — known jailbreak technique reference · informational: not an instruction file1. Settings > Apps (or Apps & Connectors) > Advanced settings: turn on **Developer mode**Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.
AST01
MEDIUM
Hidden zero-width characters detectedtests/test_config.py:48 · Hidden zero-width characters detected — may conceal content"# a comment\n\nexport A=1\nB = \"two words\"\nC='3'\nnot a setting\nD=\nE=x=y\n".encode("utf-8")Fix: Keep SKILL.md to visible, descriptive text. Remove comments and encoded text that address the agent.
AST07
MEDIUM
MCP server using network transport (SSE/HTTP)src/research_mcp/remote.py:206 · MCP server using network transport (SSE/HTTP) — data sent over networkinner = mcp_server.streamable_http_app(Fix: Namespace tool names and keep tool descriptions literal; no instructions to the agent inside them.
AST04

Cost to the agent

Instructions loaded
≈58 tok
1 file · 81 in bundle
Tool list per prompt
≈0 tok
0 tools · 0 per tool
Handshake
no reply
launch to tools/list, stdio
Undocumented tools
0
description under 20 chars · avg 0

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

7 points to 80 · weights: behavior 45%, static 25%, prompt 15%, publisher 15%
  1. Unclaimed publisher≈ +8 on the totalClaim russell-day: a verified publisher gets +50 on publisher trust. It is free and usually instant with GitHub sign-in.
  2. 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.
  3. Auto-update or dynamic fetch configuration≈ +4 on the totalPin versions and sources; vendor anything fetched at install time.
  4. Tunneling service URL≈ +4 on the totalReplace encoded or assembled strings with plain literals and remove undeclared network calls.
  5. Process kill command≈ +2 on the totalBuild commands from fixed arguments; do not interpolate untrusted strings into a shell.
  6. Python subprocess execution detected≈ +2 on the totalBuild commands from fixed arguments, never from interpolated strings; prefer an API client over shelling out.
Publish a new version to PyPI, then queue it here. Not your skill? Claim the handle to get verdict changes by email.Claim russell-dayPropose a patch to the maintainer