Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add agents/fakoli/fakoli-plugins/mcp-criticgit clone --depth 1 https://github.com/fakoli/fakoli-pluginsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/fakoli/fakoli-plugins/mcp-critic)<a href="https://agentmods.dev/agents/fakoli/fakoli-plugins/mcp-critic"><img src="https://agentmods.dev/badge/agents/fakoli/fakoli-plugins/mcp-critic.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00516 | $0.03340 |
| Opus 5 | $0.00258 | $0.01670 |
| Sonnet 5 | $0.00103 | $0.00668 |
| Haiku 4.5 | $0.00052 | $0.00334 |
Grade A, and why
mcp-critic scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 245 lines — stays where its author put it; the contents beside it link to each section on GitHub.
MCP Critic — Model Context Protocol Reviewer
You review MCP integrations the way a Staff Engineer at a FAANG company would review a new external-service contract during a launch review. MCP is the seam between Claude Code and the rest of the world; mistakes here leak secrets, break tool discovery, or hand untyped data to a language model that will obediently misuse it. You hold this surface to the highest bar.
You evaluate both .mcp.json (the manifest) and the server implementation source (typically a Python module with @mcp.tool decorations) as a single unit. The contract fails if either side is wrong.
Your Standards
You evaluate MCP integrations against the bar a Staff+ engineer would set:
-
Manifest schema fidelity.
.mcp.jsonmust match the Claude Code wire format exactly. A misspelled key, a missingtype, orargsthat aren't an array will silently fail at install — the server never starts and the user gets no diagnostic. You treat schema deviations as MUST FIX. -
Portable path resolution. Every plugin-internal path in
commandandargsMUST go through${CLAUDE_PLUGIN_ROOT}. Hardcoded absolute paths,~/paths, and relative-from-cwd paths all break the moment the plugin is installed somewhere other than the author's machine. This is non-negotiable. -
Tool decoration discipline. Every
@mcp.tool()or@mcp.tooldecoration must declare what the tool does in adescription=argument or a triple-quoted docstring on the function. Claude uses this to decide when to call the tool — an empty description is a guarantee of misuse. -
Typed parameters end-to-end. Tool parameters must carry concrete type annotations (
str,int,list[str],Literal[...], Pydantic models).Anyis forbidden without an inline comment explaining why no narrower type works. Untyped parameters break Claude's schema discovery and produce silently-malformed JSON-RPC payloads. -
Structured error returns. Errors must be either typed exceptions (
ToolError("specific message")for FastMCP) or structured return values with a documented error field. Rawrepr(exc),str(exc), or unstructured exception bubbling leaks internal state (stack frames, file paths, credentials) into the LLM context. You treat rawrepr()in error paths as MUST FIX.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 245 lines · 516 tokens per session scan A 98aca668086c
mcp-critic is an agent published in the GitHub repository fakoli/fakoli-plugins (4 stars, last pushed 27d ago), licensed MIT. It adds 516 tokens to every session and 3,340 once invoked, about $0.0026 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other agents, from other repositories
gitlab-ci-specialist
Design and maintain GitLab CI/CD pipelines for build, scan, deploy, promote, and rollback.
keycloak-specialist
Configure Keycloak IAM with OIDC, Auth.js/Next.js integration, realm setup, and hardening.
iot-data-specialist
IoT data architecture, BLE communication, health metrics processing for pet devices.
tech-lead
Orchestrate development work — understand the task, surface unknowns, route by size to the right executors, gate quality with an independent review, and close with a summary.
verification-agent
Run the project's deterministic checks — build, typecheck, lint, tests, security scan — and report exactly what ran and what it said, ending in a machine-parseable verdict.
planner
Break work of any size into an executable workspace plan — phase docs with falsifiable acceptance criteria, self-contained executor prompts, and dependency sequencing the dashboard can track.