usage-first-mcp-development

usage-first-mcp-development is a skill for Claude Code, Codex from tyler-mitchell/type-atlas. It costs 47 tokens per session (5,141 once invoked), scanned A, original, Apache-2.0.

A method for building code-understanding tools around real questions encountered while working on Type Atlas. MCP is a standard way for an AI agent to call external tools and receive structured answers.

In plain words
What is it for?
Use it when designing or testing MCP tools that help an agent investigate a codebase and decide what change to make.
Why use it?
It exposes weaknesses in a tool by using it during genuine development, so the answers can be improved where they leave an agent stuck.

Skill for Claude CodeCodex

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.

agentmods
npx agentmods add skills/tyler-mitchell/type-atlas/usage-first-mcp-development
Any agent
npx skills add tyler-mitchell/type-atlas --skill usage-first-mcp-development
Clone the repo
git clone --depth 1 https://github.com/tyler-mitchell/type-atlas

Made for: Claude Code, Codex.

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 usage-first-mcp-development

README.md
[![agentmods](https://agentmods.dev/badge/skills/tyler-mitchell/type-atlas/usage-first-mcp-development.svg)](https://agentmods.dev/skills/tyler-mitchell/type-atlas/usage-first-mcp-development)
Your own site
<a href="https://agentmods.dev/skills/tyler-mitchell/type-atlas/usage-first-mcp-development"><img src="https://agentmods.dev/badge/skills/tyler-mitchell/type-atlas/usage-first-mcp-development.svg" alt="Measured on agentmods" height="20"></a>
Per session 47 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 5,141 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
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 $0.00047 $0.05141
Opus 5 $0.00023 $0.02570
Sonnet 5 $0.00009 $0.01028
Haiku 4.5 $0.00005 $0.00514

Measured 5d ago against content hash f2631cff0497, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-05, from the pricing page.

Security

Grade A, and why

usage-first-mcp-development 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.

.skills/usage-first-mcp-development/SKILL.md · 400 lines

How it starts

The opening of the file, as written. The whole thing — 400 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Usage-First MCP Development

The work

You are trying to understand a codebase. Really trying — there is something you need to know to make the next change correctly, and you do not know it yet.

The instrument you use to find out is the one you are building.

That is the whole method. Not a process to execute: a situation to be in. Do real work, need real answers, and pay attention to what happens when you ask.

An agent that treats this as a task sheet produces exactly what a task sheet produces — locally valid changes, no design, and a tool that nobody can think with. The rails at the end of this document are cheap insurance against specific expensive mistakes. They are not the point, they are not the shape of the work, and narrating them is worse than useless. Follow them invisibly.


What "good" means here

The product is not a function call. It is the answer an agent receives and has to act on.

So the bar is ergonomic, and it is answered by imagining a competent agent that did not write any of this:

They asked a question because they were stuck. Does this answer unstick them, or do they now have to open a file, ask again with different arguments, or guess?

An answer that is correct and leaves them stuck has failed. Every reformulation they are forced into is a defect you caused.

Everywhere else you have worked, your tools were trustworthy and the code was suspect. Here the tool is the code, and a malformed answer does not announce itself — it arrives looking like an answer and you act on it. Which is why: source, types, and tests say what the code says; only a call says what an agent receives. Claims about behaviour need a call behind them.

One boundary on that rule, learned the hard way. It governs the MCP surface — what an agent receives. For the substrate beneath it — Volar, TypeScript, the bridge — the repo's recorded evidence outranks your live probe: docs/volar-affordance-evidence.md says what was true, and instrumentation only says what changed. A substrate diagnosis built from probes alone was confidently wrong here for ten days, while the observation contradicting it sat unread in the ledger the whole time. Read the ledger first, and when it disagrees with a live call, you have found a change, not a refutation.

Read the full file on GitHub · 400 lines

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. 5d ago First seen · 400 lines · 47 tokens per session scan A f2631cff0497

Subscribe to this mod's changes

usage-first-mcp-development is a skill published in the GitHub repository tyler-mitchell/type-atlas (2 stars, last pushed 7d ago), licensed Apache-2.0. It adds 47 tokens to every session and 5,141 once invoked, about $0.0002 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.

Related

Other skills, from other repositories

ragcode-agents

Orchestrate multi-agent workflows through RagCode. Turn a natural-language collaboration pattern (e.g. "Gemini plans, Claude breaks down, Codex implements, Claude reviews") into a validated DAG workflow spec, then run it with the RagCode agent workflow engine. Trigger when the user asks to coordinate multiple AI…

MarshallEriksen-Neura/ragcode · 96 tokens

ragcode-memory

Manage shared cross-AI memories for this project through RagCode MCP tools (memorywrite, memoryquery, memorylist, memorydelete), with ragcode memory CLI fallback when MCP is unavailable. Read prior decisions, user preferences, and feedback written by other AI agents, and record new insights so they persist across…

MarshallEriksen-Neura/ragcode · 101 tokens

idesense

Guide for using JetBrains IdeSense tools for code navigation and analysis (plus the retained rename/move refactorings). TRIGGER: When ANY of these MCP tools are available in the current session: idefindusages, idefinddefinition, idefindclass, idefindfile, idesearchtext, idediagnostics, ideexplainsymbol…

vcth4nh/idesense · 172 tokens

codeindex

Answer code questions structurally instead of reading whole files. Use when you need to find where something is defined, read one function, see what calls a symbol, trace imports, judge what a change breaks, or orient yourself in an unfamiliar file or repository. Reach for this before Read, Grep or Glob on a code…

munhq/codeindex · 148 tokens

opik-diagnose

Surface the Opik traces worth a developer's attention, ranked by signal — errors, failed tool calls, latency, regressions, and low online-eval scores — plus Diagnostics issues. Reads live/production traces via the SDK (searchtraces and agentinsights) and works with no MCP; uses the MCP issue entity when connected.…

comet-ml/opik-mcp · 147 tokens

lsp-implement

Find all concrete implementations of an interface or abstract type. Use when you need to know what types satisfy an interface, or what subtypes exist before changing a base type.

blackwell-systems/agent-lsp · 39 tokens