Skip to main content
Glama

MCP Preflight

Review MCP config before you run it.

Turn a configuration or tool-catalog JSON snapshot into a static review report. The checker never starts or contacts the configured servers. The Python CLI has no third-party runtime dependencies and needs no account or API key.

CI

Alpha, MIT licensed. No PyPI distribution; use a reviewed GitHub release or authorized checkout. CLI and optional MCP stdio adapter are the focus. A bounded Windows socket-close fix addresses the reproduced local HTTP timeout; see scope and tests. A clean report is not proof of safety.

Try the difference

From an authorized checkout, with Python 3.12+:

python scripts/mcp_preflight.py --input examples/preflight-before.json --fail-on review
python scripts/mcp_preflight.py --input examples/preflight-after.json --fail-on review

Run the commands separately: the first intentionally exits 1; the second exits 0. No pip install, server startup or network access is needed for this demo.

Synthetic example: an unpinned package selector triggers a finding; an exact version removes that finding. Runtime remains unverified.

The only input change is example-mcp@latest → example-mcp@1.2.3. Both are synthetic strings, not package recommendations. The checker does not look up whether that package or version exists. Both inputs explicitly provide an empty catalog; this demo checks version syntax, not tool safety.

Related MCP server: Argus

Use it in your workflow

You have…

Preflight gives you…

An MCP config someone wants to enable

Static indicators to review before connection

A captured tools/list response

Catalog/schema and annotation review items

A CI check

JSON or SARIF plus a configurable failure threshold

Codex with the optional adapter

preflight_review_json over stdio

python scripts/mcp_preflight.py --input your-config.json --output report.json --fail-on high
python scripts/mcp_preflight.py --input your-config.json --format sarif --output report.sarif

Reports omit supplied credential values, URLs, commands and free-form names and descriptions. Keep the original input private and inspect any report before sharing. SARIF export works; GitHub code-scanning ingestion is not yet verified.

Next: input examples · CLI reference · Codex setup · complete usage

Where it fits

Preflight is a small static review step. For agent-wide discovery, broader security analysis or live protocol debugging, compare Snyk Agent Scan, Cisco MCP Scanner, SecureAI-Scan and MCP Inspector. Some also offer offline scanning: that is not an exclusive feature here. See the dated comparison and tradeoffs.

Preflight does not inspect server source code, fetch catalogs, enforce runtime policies or certify prompt-injection resistance. Detection accuracy and competitive superiority have not been benchmarked.

Evidence you can inspect

Remote CI on 2026-10-06: 13 successful jobs covering one artifact build, four Windows/Linux × Python 3.12/3.13 core jobs, and eight MCP jobs with minimum/normal SDK resolution. The workflow checks installed CLI/HTTP and MCP behavior outside the checkout. This is evidence for that commit, not a guarantee for future changes.

Validation and open limits · HTTP behavior · Contributing · Security reporting

Help improve the alpha

Authorized reviewers: report a reproducible false positive or missing check. Use a minimal synthetic example and include the rule ID and expected behavior. Never attach live credentials or original private configurations.

Licensed under MIT. Public availability, customer adoption and a support SLA are not asserted.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      Enables agents and CI pipelines to audit MCP servers and agent tool-chains by statically scanning repositories, local checkouts, tools/list exports, or live endpoints for risks such as destructive actions without confirmation, mismatched safety annotations, injection surfaces, credential or PII exposure, and unguarded command, path, or URL sinks. All checks are read-only and never execute the scanned code or call tools/call.
      3
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Runs read-only, offline static scans of the local machine's MCP server configs and agent skill folders, surfacing prompt injection, hidden Unicode, dangerous commands, credential paths, hardcoded secrets, unpinned packages, and over-broad filesystem or transport scope as text, JSON, SARIF, or HTML reports an agent can act on. Also allows scanning a single downloaded skill folder or a saved tools/list JSON so an agent can vet third-party components before installing them.
      47 npm
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Lets an agent inspect and audit MCP servers before connecting to them, checking their exposed tools, prompts and resources for tool poisoning, shadowing, excessive permissions, leaked secrets and unsafe transports against 24 deterministic rules. Results are returned as findings with evidence, remediation and CWE/OWASP references, and can fail CI with SARIF, Markdown or standalone HTML reports.
      Apache 2.0
    • A
      license
      A
      quality
      B
      maintenance
      Enables static, read-only auditing of MCP servers’ registry entries, package metadata, client configurations, and tool definitions for supply-chain security indicators. It returns merged reports with transparent skipped checks and never executes the inspected code.
      6
      113 PyPI
      MIT