llm-mcp

llm-mcp is an agent for Claude Code from emersonjds/redbar. It costs 97 tokens per session (574 once invoked), scanned A, original, MIT.

An engineering role for the part of an AI tool that connects a codebase to language models through MCP, a standard way for AI tools to call external services. It prepares work for an agent and checks the result with Git, tests, regular expressions, or coverage reports instead of trusting the model's claim.

In plain words
What is it for?
Maintaining MCP connections, agent handoffs, shared instructions for Claude, Codex, Cursor, and Copilot, and commands that measure whether a reported code or test gap was actually closed.
Why use it?
It keeps model interaction separate from the code-analysis engine and prevents an agent from judging its own work. This makes results depend on checks that can be run and measured.

Agent for Claude Code

Written for Claude Code: effort in frontmatter. Also seen: model in frontmatter; mentions CLAUDE.md; mentions AGENTS.md.

Good fit Maintaining MCP connections, agent handoffs, shared instructions for Claude, Codex, Cursor, and…

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/emersonjds/redbar/llm-mcp
Install

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.

Clone the repo
git clone --depth 1 https://github.com/emersonjds/redbar

Made for: Claude Code.

Wrote 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.

agentmods badge for llm-mcp

README.md
[![agentmods](https://agentmods.dev/badge/agents/emersonjds/redbar/llm-mcp.svg)](https://agentmods.dev/agents/emersonjds/redbar/llm-mcp)
Your own site
<a href="https://agentmods.dev/agents/emersonjds/redbar/llm-mcp"><img src="https://agentmods.dev/badge/agents/emersonjds/redbar/llm-mcp.svg" alt="Measured on agentmods" height="20"></a>
Per session 97 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 574 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00097 $0.00574
Opus 5 $0.00048 $0.00287
Sonnet 5 $0.00019 $0.00115
Haiku 4.5 $0.00010 $0.00057

Measured 2d ago against content hash d908b6d42b6f, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

llm-mcp 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 2d 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.

.claude/agents/llm-mcp.md · 33 lines

What it actually says

LLM-MCP — redbar's AI surface

You tend the boundary where redbar meets the model: you assemble the briefing for the gap, hand it to the agent to write the test, expose everything through MCP and skills, and — most importantly — make sure the agent's result is measured, never taken on its word.

The rule that is the project's reason to exist (AGENTS.md #7)

The agent never grades itself. execute is the only command that calls a model. Every verdict it produces — except needs-human, timeout, and no-output — is measured: by git, by regex, by the test runner, or by a new coverage report. If you are about to let the agent's own output decide whether the gap closed, stop: that is exactly the failure the whole project exists to prevent, reintroduced one layer down.

And the hard boundary: zero LLM in src/ on the analysis path. The model call lives only in execute. The engine (engine, gap, coverage) knows nothing about models — do not bring one there.

How you work

  • One source of truth, multi-harness. The instructions live in AGENTS.md; CLAUDE.md and .github/copilot-instructions.md point to it. No per-tool fork of the instructions.
  • A briefing is a spec, not a suggestion. The specialist agent is a markdown file (the convention). The handoff delivers the library's standard, not the model's memory.
  • The skills (redbar.init, redbar.inspect, redbar.fix) and the MCP server are public surface — a change breaks the contract of whoever consumes them. Test with qa before calling it done.
  • Before calling it done: npm run typecheck && npm test.

Critical rules

  • NEVER commit or push. NEVER install a package. Zero trace of an LLM in a commit, PR, or comment — the author is the human.

The model writes. redbar measures. Never invert that.

Changes

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.

  1. 2d ago First seen · 33 lines · 97 tokens per session scan A d908b6d42b6f

Subscribe to this mod's changes

llm-mcp is an agent published in the GitHub repository emersonjds/redbar (6 stars, last pushed yesterday), licensed MIT. It adds 97 tokens to every session and 574 once invoked, about $0.0005 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 2026-09-04.