MCP Scorecard v1.0.0

MCP Scorecard v1.0.0

v1.0.0 is the first stable release of MCP Scorecard.

This release takes the existing deterministic scanner and hardens it into a release-grade
CI-first scorecard for MCP servers. The core philosophy stays the same: local discovery,
deterministic checks, stable scoring, and machine-readable output. What changes in v1.0.0 is the
contract quality, naming consistency, and release surface.

Highlights

• stable V1 JSON scorecard report contract
• explicit score buckets:
  • conformance
  • security
  • ergonomics
  • metadata
• explicit check metadata:
  • id
  • title
  • bucket
  • severity
  • rationale
• scorecard-oriented GitHub Action outputs and PR summary
• preferred CLI name mcp-scorecard, with mcp-trust retained as a compatibility alias

Included In v1.0.0

• JSON reports now expose a stable top-level contract:
  • schema
  • generator
  • scan
  • inventory
  • scorecard
  • checks
  • findings
  • grouped_findings
  • metadata
• terminal output now mirrors the scorecard contract:
  • generator
  • schema version
  • scan timestamp
  • target description
  • total score
  • category scores
  • findings by bucket
  • limitations
• SARIF now carries aligned scorecard metadata in run properties and result properties
• sample reports regenerated from the current scanner
• release docs, checklist, and README updated for v1.0.0

Compatibility Notes

• the repository and action reference are now aak204/MCP-Scorecard
• mcp-scorecard is the preferred CLI
• mcp-trust remains available as a compatibility alias
• the Python module remains mcp_trust

Score Meaning

• a high score means fewer deterministic findings, not a guarantee of safety
• a low score means more deterministic findings or higher-risk exposed surface, not malicious intent
• the score measures deterministic, reviewable properties only

Quickstart

Local:

python -m venv .venv
source .venv/bin/activate
pip install -e .[dev]
mcp-scorecard scan --json-out mcp-scorecard-report.json --sarif mcp-scorecard-report.sarif --cmd python examples/insecure-server/server.py

GitHub Actions:

- name: Run MCP Scorecard
  id: scorecard
  uses: aak204/MCP-Scorecard@v1.0.0
  with:
    cmd: python path/to/your/server.py
    min-score: "80"
    json-out: mcp-scorecard-report.json
    sarif-out: mcp-scorecard-report.sarif
    markdown-out: mcp-scorecard-summary.md

Validation Snapshot

• examples/insecure-server -> 10/100
• @modelcontextprotocol/server-memory@2026.1.26 -> 100/100
• @modelcontextprotocol/server-filesystem@2026.1.14 -> 40/100

Out Of Scope

• LLM scoring
• hosted scanning
• registry integration
• certification-style claims

MCP Trust Kit v0.5.0

MCP Trust Kit v0.5.0

v0.5.0 is a narrow integration-driven release.

The scanner contract from v0.4.0 stays intentionally stable: local stdio discovery, deterministic rules, predictable scoring, terminal summary, JSON, SARIF, and GitHub Actions. The main reason for v0.5.0 is to make Layer 1 baseline output easier to consume by downstream systems that care about scan freshness and temporal decay.

Highlights

• explicit scan_timestamp field in JSON output
• matching timestamp metadata in SARIF
• no break to existing generated_at consumers
• release-ready static baseline contract for higher-layer integrations

Included In v0.5.0

• JSON reports now expose:
  • scan_timestamp
  • generated_at
  • aggregate score breakdown
  • capability-aware and hygiene-aware findings
• SARIF runs now expose:
  • runs[].properties.scan_timestamp
  • runs[].invocations[].endTimeUtc
• sample reports regenerated from the current scanner
• release docs and README updated for v0.5.0

Validation Snapshot

• examples/insecure-server -> 10/100
• @modelcontextprotocol/server-memory@2026.1.26 -> 100/100
• @modelcontextprotocol/server-filesystem@2026.1.14 -> 40/100

Contract Note

generated_at is still present for backward compatibility.

scan_timestamp is now the canonical timestamp field for downstream integrations that need to reason about baseline freshness.

Quickstart

python -m venv .venv
source .venv/bin/activate
pip install -e .[dev]
mcp-trust scan --json-out baseline.json --cmd python examples/insecure-server/server.py

GitHub Actions:

- name: Run MCP Trust Kit
  uses: aak204/MCP-Trust-Kit@v0.5.0
  with:
    cmd: python path/to/your/server.py
    min-score: "80"
    json-out: mcp-trust-report.json
    sarif-out: mcp-trust-report.sarif

MCP Trust Kit v0.4.0: First Public Release

MCP Trust Kit v0.4.0

v0.4.0 is the first practically useful public release of MCP Trust Kit.

This release keeps the product intentionally narrow: local stdio MCP discovery, deterministic
rules, predictable scoring, terminal summary, JSON, SARIF, and GitHub Actions. The main change is
not "more features for the sake of it". The main change is that the score now behaves more like a
review signal and less like a demo number.

Highlights

• deterministic surface-risk scoring for MCP servers
• refined schema heuristics that stop penalizing empty object schemas for no-arg tools by default
• expanded rules for schema hygiene and risky exposed capabilities
• capability-aware report summaries
• terminal output that explains why the score is low and which tools to review first
• real-world validation against public MCP servers
• Bash-first quickstart and production-ish GitHub Action docs

Included In v0.4.0

• local stdio discovery transport
• normalized server, tool, finding, and report models
• deterministic rule set covering:
  • metadata hygiene
  • schema hygiene
  • command execution
  • filesystem mutation
  • network request surface
  • download-and-execute patterns
• penalty-based score breakdown across spec, auth, secrets, and tool_surface
• terminal summary, JSON report, and SARIF export
• demo insecure MCP server
• sample reports, validation docs, and CI workflow

Validation Snapshot

• examples/insecure-server -> 10/100
• @modelcontextprotocol/server-memory@2026.1.26 -> 100/100
• @modelcontextprotocol/server-filesystem@2026.1.14 -> 40/100

The key interpretation stays the same:

• low score means higher exposed surface risk, not malicious intent
• high score means fewer deterministic findings, not a guarantee of safety

Quickstart

Local:

python -m venv .venv
source .venv/bin/activate
pip install -e .[dev]
mcp-trust scan --cmd python examples/insecure-server/server.py

GitHub Actions:

- name: Run MCP Trust Kit
  uses: aak204/MCP-Trust-Kit@v0.4.0
  with:
    cmd: python path/to/your/server.py
    min-score: "80"
    json-out: mcp-trust-report.json
    sarif-out: mcp-trust-report.sarif