Results · October 2026 · Sealmark 0.2.1
What Sealmark found on 30 public MCP servers
Each server was installed the way its README says, run through Sealmark with the standard settings, and every verdict was then checked independently by reading the source and making direct calls. Open a server to see each check, what it measured and what was not covered.
| Server | Version | Protocol | Deeper packs | Coverage |
|---|---|---|---|---|
| @modelcontextprotocol/server-memorynpm | 1.0.0 | A | none applied | low |
| @modelcontextprotocol/server-sequential-thinkingnpm | 1.0.0 | A | none applied | hygiene only |
| @openbnb/mcp-server-airbnbnpm | 0.3.0 | A | none applied | hygiene only |
| @upstash/context7-mcpnpm | 4.4.1 | A | none applied | hygiene only |
| arxiv-mcp-serverPyPI | 0.8.2 | A | none applied | low |
| duckduckgo-mcp-serverPyPI | 0.7.0 | A | A Network | low |
| git-mcp-serverPyPI | 0.2.5 | A | none applied | hygiene only |
| mcp-ripgrepnpm | 0.4.0 | A | none applied | hygiene only |
| mcp-server-duckdbPyPI | 1.1.0 | A | none applied | hygiene only |
| mcp-server-fetchPyPI | 2026.10.10 | A | A Network | covered |
| mcp-server-gitPyPI | 2026.10.10 | A | A GitA Effects | covered |
| mcp-server-motherduckPyPI | 1.1.0 | A | none applied | low |
| mcp-server-timePyPI | 2026.10.10 | A | none applied | hygiene only |
| mcp-text-editorPyPI | 1.2.2 | A | none applied | hygiene only |
| @itseasy21/mcp-knowledge-graphnpm | 1.0.7 | B | none applied | hygiene only |
| @modelcontextprotocol/server-filesystemnpm | 1.0.0 | A | B FilesA Effects | covered |
| @executeautomation/database-servernpm | 1.1.0 | B | C SQLite | low |
| @kazuph/mcp-fetchnpm | 1.6.3 | A | D Network | covered |
| @modelcontextprotocol/server-postgresnpm | 0.6.2 | A | D Database | covered |
| mcp-server-sqlitePyPI | 2025.4.25 | A | D SQLite | low |
| @agent-infra/mcp-server-filesystemnpm | 1.2.29 | A | F Files | covered |
| @bytebase/dbhubnpm | 1.5.0 | A | F Database | covered |
| @cyanheads/filesystem-mcp-servernpm | 1.0.4 | A | F Files | covered |
| @cyanheads/git-mcp-servernpm | 2.15.3 | A | D GitF NetworkA Effects | covered |
| @henkey/postgres-mcp-servernpm | 1.0.7 | A | F Database | low |
| @modelcontextprotocol/server-everythingnpm | 1.0.0 | D | F Network | low |
| fetch-mcpnpm | 0.0.5 | A | F Network | covered |
| markitdown-mcpPyPI | 0.0.1a7 | A | F Network | covered |
| mcp-sqlitenpm | 1.0.9 | A | F SQLite | low |
| postgres-mcpPyPI | 0.3.0 | A | F Database | low |
30 of 30 servers. A * means a critical check could not be judged, so the grade is provisional.
How to read a grade
Protocol only means hygiene
The protocol pack calls tools only with invalid input, so a server that does little scores well there. Look at the deeper packs.
An A is not an endorsement
It means these probes passed for one version and configuration; every page lists what was not covered. See how grades work and what an A means.
How reliable is Sealmark itself?
Every verdict on the 30 servers was checked independently (reading the source and making direct calls). The tallies include deliberately broken test servers the verifiers wrote, which is where most wrong verdicts come from. Sealmark makes mistakes. Here is every verdict that was checked against ground truth, so you can see how often it was right.
| Run | Correct pass | Correct fail | Wrong pass | Wrong fail | Inconclusive | Not applicable |
|---|---|---|---|---|---|---|
| First 10 servers, third pass | 127 | 75 | 12 | 2 | 23 | 31 |
| 20 new servers | 130 | 82 | 27 | 4 | 28 | 63 |
| Total | 257 | 157 | 39 | 6 | 51 | 94 |
Most wrong passes came from servers the verifiers deliberately broke to test whether Sealmark would catch them. Every one found became a fix. The full method is in the methodology.