How We Evaluate MCP Tools
The trust checklist behind MCP recommendations on agentk.it.
MCP tools need a stronger review process than normal link directories because they can connect Agents to real systems.
The minimum bar
For GitHub-discovered tools, agentk.it uses a star floor for discovery. Stars are not enough, but they reduce obvious low-signal noise.
After discovery, the tool still needs review:
- What does it actually help with?
- Is the source official or community-maintained?
- Is the license clear?
- Has the repository been updated recently?
- Does it document the Agents it supports?
- What data or permissions does it need?
What should appear publicly
The public page should focus on the value and trust signals:
- What the tool is.
- What it helps the Agent do.
- Which Agent, Skill, MCP, or Workflow scenarios it fits.
- Official website, GitHub, docs, quick start, and release links.
- Stars, license, latest update, and last checked date.
What should stay internal
Raw extracted install snippets, guessed compatibility, noisy README fragments, and crawler uncertainty belong in the candidate pool. They help reviewers, but they should not be treated as public guidance.