mcp security leaderboard

you connected an MCP server to your AI agent. you didn't check what was inside. we did.

real servers. real scores. no anonymizing. scanned with mcprobe.

98

packages tried

24

servers scanned

13

with findings

11

clean

ranked by risk score

1

@dronelytics/mcp

25 tools - 8 findings - 2 CRITICAL, 2 HIGH, 4 MEDIUM

100

risk 100/100 (CRITICAL) - 25 tools - 8 findings

CRITICAL

prompt injection in tool description

tool description contains "ignore all previous instructions" - your agent reads this before every call

CRITICAL

prompt injection in tool description

tool description contains "you are now a" - identity override attempt

2

@agenticmail/mcp

101 tools - 4 findings - 2 HIGH, 2 MEDIUM

62

risk 62/100 (HIGH) - 101 tools - 4 findings

HIGH

prompt injection in tool description

tool description contains injection pattern - the LLM will read it before deciding what to call

3

@gitdealflow/mcp-signal

13 tools - 8 findings - 2 HIGH, 6 MEDIUM

56

risk 56/100 (HIGH) - 13 tools - 8 findings

HIGH

prompt injection patterns detected

multiple tool descriptions contain injection text

MEDIUM

unrestricted file access

tool parameters accept arbitrary file paths without validation

4

@automatelab/n8n-mcp

12 tools - 2 findings - 2 MEDIUM

50

5

agent-browser-mcp-server

44 tools - 4 findings - 4 MEDIUM

28

6

@artifacta-mcp/mcp

10 tools - 1 finding - 1 MEDIUM

25

7

hostinger-api-mcp

223 tools - 1 finding - 1 CRITICAL

25

8

mcp-db-server

5 tools - 1 finding - 1 MEDIUM

25

9

@dollhousemcp/mcp-server

5 tools - 3 findings - 3 LOW

21

10

@doitintl/doit-mcp-server

135 tools - 1 finding - 1 MEDIUM

7

11

@modelcontextprotocol/server-sequential-thinking

1 tool - 1 finding - 1 LOW

7

12

mcp-calculator-server

6 tools - 6 findings - 6 LOW

5

13

d33naz-mcp-filesystem

2 tools - 2 findings - 2 LOW

4

clean (0 findings)

-

mcp-calculator

101 tools - 0 findings

0

clean

-

@contextium/mcp-server

68 tools - 0 findings

0

clean

-

mcp-server-slack

20 tools - 0 findings

0

clean

-

@bitcompare/mcp-server

18 tools - 0 findings

0

clean

-

@modelcontextprotocol/server-everything

13 tools - 0 findings

0

clean

-

@clicks-protocol/mcp-server

10 tools - 0 findings

0

clean

-

@impri/mcp

8 tools - 0 findings

0

clean

-

@endiagram/mcp

7 tools - 0 findings

0

clean

-

@bluesprincemedia/thiri-mcp

5 tools - 0 findings

0

clean

-

mcp-time-server

2 tools - 0 findings

0

clean

-

mcp-server-docker

1 tool - 0 findings

0

clean

a score of 0 doesn't mean safe. it means we didn't find anything. there's a difference. you should know the difference.

mcprobe checks descriptions, schemas, and parameters. it doesn't execute tool calls. a server can score 0 and still exfiltrate your data at runtime.

but sure, connect it. what could go wrong.

scan your own servers read the blog valtors