Agent Radar

io.github.zationPublished by GitHub account zationv0.9.2Updated Oct 25, 2026

Search, inspect, recommend, and explain rated AI tools through Agent Radar.

VerifiedStreamable HTTPWeb executableKnowledge & MemoryDeveloper Tools

Overview

AI-generated overview

A remote MCP server for searching, inspecting, recommending and explaining rated AI tools from the Agent Radar catalog.

What you can do
Search the reviewed catalog of AI agents, skills, MCP servers, CLIs, frameworks and prompts by query text, type, tag or risk filters, and open one Tool Card by its stable ID with its rating result. Get task-oriented recommendations with fit reasons, risks, alternatives and a conservative next action such as use, compare, ask_human or avoid. Explain a tool's rating through its dimensions, penalties and boosts. All four tools are read-only with respect to installing or running tools.
When to use it
Useful when choosing among AI agents, skills, MCP servers, CLIs or frameworks for a concrete task and you want reasons, permissions, risk signals and traceable evidence rather than popularity rankings. Also suited to coding agents that need an agent-readable decision layer over the AI tooling ecosystem.
Requirements
A remote Streamable HTTP endpoint; nothing to install locally. The optional secret header X-Agent-Radar-LLM-API-Key supplies a per-request LLM provider API key for recommend_tools; without it the server may use its own configured fallback. Network access to the endpoint is required.
Before you install
The LLM provider API key sent in X-Agent-Radar-LLM-API-Key is a credential you supply per request, so only send it to an endpoint you trust. Recommendations are advisory: the server does not install, authorize or run third-party tools and is not a substitute for a security scanner, dependency audit or human review. Ratings and recommendations can be lowered by unknown, stale or conflicting evidence.

Installation

In SourceWeft

  1. Open Agent Radar in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Use with an agent

Paste this prompt into Claude Code, Codex, Cursor or another agent — or into SourceWeft chat. The agent reads this server's install guide, shows you its publisher, endpoint and sign-in, and adds it to your client once you agree. Keys never go into a command.

Read https://sourceweft.com/mcp/io.github.zation%2Fagent-radar/install.md and connect the MCP server it describes to my client. Before changing any config, show me its publisher, endpoint and sign-in method and wait for my OK. Never put an API key in a command or in a file I would commit; ask me where it should come from.

Read the install guide the agent follows

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "agent-radar": {
      "type": "http",
      "url": "https://agent-radar.zation1.workers.dev/api/mcp"
    }
  }
}

README

[Agent Radar]

Agent Radar

Choose AI tools with evidence, not hype.

Agent Radar is a structured rating and recommendation knowledge base for AI Agents, Skills, MCP Servers, CLIs, Frameworks, and Prompts/Rules. It helps developers and coding agents answer a practical question: which tool fits this task, and what should I know before trusting it?

Instead of publishing another directory of links, Agent Radar turns fragmented ecosystem information into Tool Cards, explainable ratings, task-oriented recommendations, risk signals, and agent-readable data.

Why Agent Radar

The AI tooling ecosystem is growing faster than any one developer can evaluate it. Names and categories overlap, documentation quality varies, and popularity often says little about task fit, integration cost, maintenance, permissions, or trust.

Agent Radar adds a decision layer:

  • Task fit over popularity. Recommendations start from the work to be done, not a global leaderboard.
  • Evidence over marketing. Sources, update times, provenance, and confidence remain visible.
  • Explanation over opaque ranking. Ratings and recommendations expose reasons, trade-offs, and unsuitable conditions.
  • Safety before execution. Permission, trust, data-flow, and production-impact risks can tighten or block a recommendation.
  • Built for humans and agents. The same reviewed knowledge is available through a Web UI, static data, HTTP endpoints, and MCP.

What You Get

Structured Tool Knowledge

Tool Cards describe capabilities, appropriate tasks, limitations, integration methods, maintenance, permissions, known risks, sources, and confidence in a consistent format across tool types.

Explainable Ratings

Ratings combine task fit, maintenance, documentation, integration cost, security risk, and evidence quality. Scores never replace the underlying evidence or risk context.

Task-Oriented Recommendations

Describe a development task, environment, preferred tool type, allowed permissions, and risk tolerance. Agent Radar returns candidates with fit reasons, risks, alternatives, and a conservative next action.

ActionMeaning
useA suitable option can be included in the plan within its stated boundaries
compareMultiple options deserve a trade-off review
ask_humanPermission, account, production, or uncertainty boundaries require confirmation
avoidThe option or operation crosses an unacceptable safety or quality boundary
no_reliable_matchThe catalog cannot support a trustworthy recommendation

Agent-Readable Decision Context

JSON, JSONL, HTTP, and MCP surfaces let coding agents search the catalog, inspect Tool Cards, request recommendations, and explain ratings without scraping a human-only page.

How It Works

text
reviewed public sources  -> normalized Tool Cards and provenance  -> explainable ratings and risk signals  -> task-oriented retrieval and recommendation  -> deterministic safety enforcement  -> provider-backed evaluation and release gates  -> Web, JSON/JSONL, HTTP API, and MCP

Collection is controlled by a Source Registry. Candidates pass normalization, validation, review, admission, and promotion gates before entering reliable recommendation data. Dynamic recommendations use an LLM for task interpretation and candidate selection, while local deterministic logic validates known tools and enforces safety boundaries.

The v0.7 data path also discovers the current top two public repositories for GitHub topic agent-skills and expands eligible skills/**/SKILL.md manifests into one Tool Card per Skill. Repository rank is discovery evidence, not quality evidence. Each current build keeps only successfully fetched Skill manifests; raw Skill text stays inside ingestion evidence and is never sent to the recommendation provider.

Who It Is For

  • AI-first developers choosing among agents, Skills, MCP Servers, CLIs, and Frameworks for a concrete task.
  • Coding-agent users who want tool selection to include reasons, permissions, risks, and traceable evidence.
  • AI platform teams building a trusted, reusable catalog and decision layer for developers.
  • Tool maintainers who want capabilities, limitations, and sources represented accurately.

Trust and Safety

Agent Radar is designed to support safer selection, not to authorize execution.

  • Sources and field-level evidence remain traceable.
  • Unknown, stale, or conflicting evidence lowers confidence.
  • Permissions and data flows stay visible beside recommendation value.
  • High-risk operations can require human confirmation even when a tool is a strong task match.
  • Unknown-source code execution is treated conservatively and cannot be promoted by provider output.
  • Critical safety cases block a reviewed release when they fail, are missing, or are not executed.

Agent Radar does not automatically install, authorize, or run third-party tools. It is not a substitute for a security scanner, dependency audit, organizational policy, or human review of production-impacting actions.

Ways to Use Agent Radar

Web UI

Browse and filter the reviewed catalog, describe a task, inspect recommendations, open detailed Tool Cards, and review evaluation evidence through the Tools and Evaluation workspaces.

Static Data

Reviewed bundles expose versioned Tool Cards, Rating Results, search indexes, Golden Queries, evaluation summaries, manifests, provenance, and quality evidence as JSON or JSONL artifacts.

HTTP API

The Worker exposes read-oriented endpoints for:

  • /api/search_tools
  • /api/get_tool_card
  • /api/recommend_tools
  • /api/explain_rating
  • /api/version

Installable Agent Skill

Install the repository-owned agent-radar Skill with the open Agent Skills CLI:

bash
npx skills add zation/agent-radar --skill agent-radar

For Codex, you can also ask $skill-installer to install the Skill directly from:

text
https://github.com/zation/agent-radar/tree/main/skills/agent-radar

The Skill is local-first. Run its explicit sync command to download the latest compatible reviewed Tool Cards, Ratings, and Search Index into a versioned cache. The client verifies the v1 channel, release manifest, file sizes, SHA-256 checksums, and record schemas before atomically switching the active local release. Search, Tool Card inspection, rating explanation, and recommendation-context construction then run without MCP, without a provider key, and without transmitting the task. A failed update keeps the last verified release active. When invoked, the Skill announces that Agent Radar is in use; every successful client result carries source: "agent-radar-skill", and every Agent Radar-backed answer ends with the verified release ID and data version. Commit SHA remains available in the structured result for audit.

MCP

/api/mcp provides a stateless Streamable HTTP MCP interface built with the official TypeScript SDK v2 beta. It exposes search_tools, get_tool_card, recommend_tools, and explain_rating; all four tools are read-only with respect to tool execution and installation. /api/mcp_manifest exposes the same shared contracts for simpler HTTP integrations.

recommend_tools accepts its ordinary query and optional model as tool input. A per-request provider credential is sent only in the optional secret header X-Agent-Radar-LLM-API-Key; it is never part of the tool schema or response. The Worker may use its configured server-side fallback when that header is absent.

The remote-only Agent Radar MCP server is published in the official Registry as io.github.zation/agent-radar, backed by the production Streamable HTTP endpoint and evidence-bound GitHub OIDC publication. You can connect to and use it through either of these listings:

Production completion requires release identity convergence before functional smoke: bounded polling must observe the exact immutable tag and commit SHA from /api/version, then MCP initialize.serverInfo.version must equal the SemVer derived from that tag. The same gate runs again before Registry publication, so a previous edge version cannot pass as the new release.

Documentation

Development and Contributing

Use the Development Guide for local setup, development commands, data generation, evaluation, testing, and troubleshooting.

Coding agents and contributors must also read AGENTS.md before changing the project. It defines document authority, required validation, safe automatic actions, and changes that require human confirmation.

Source: README.md at commit 6c91d68

Tools

4

search_tools

Search Agent Radar tools

Search Agent Radar Tool Cards by query text and optional type, tag, or risk filters.

Parametersquerytop_kfilters

get_tool_card

Get an Agent Radar tool card

Return one Tool Card and its Rating Result by stable tool_id.

Parameterstool_idrequired

recommend_tools

Recommend tools

Recommend known tools for a task using request-scoped LLM credentials when required.

Parameterstaskrequiredmodeltop_kbudgetenvironmentoutput_formatexisting_toolsrisk_tolerancelanguage_or_stackallowed_permissionspreferred_tool_types

explain_rating

Explain a tool rating

Return rating dimensions, explanations, penalties, and boosts for one tool_id.

Parameterstool_idrequired

Version history

1
  1. v0.9.2LatestSep 16, 2026

Score

90% of the applicable checks met

How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.

  • Reached in the last 30 daysRequired25 pts
  • Auth declaredRequired10 pts
  • Tools listedRequired15 pts
  • DescriptionRequired15 pts
  • Logo10 pts
  • Documentation15 pts
  • License10 pts

    Declare a license in the repository or the registry entry.