Skill vs MCP
The Codex Docs MCP lets your AI search the documentation. The skill gives it direct knowledge of the API surface so it can generate queries immediately.| Codex Skill | Codex Docs MCP | |
|---|---|---|
| How it works | Preloaded API knowledge — operations, auth, templates | Searches docs on demand via MCP protocol |
| Best for | Generating runnable queries fast | Looking up specific fields, types, or guides |
| Needs network? | No — works offline | Yes — queries the MCP server |
What’s included
The skill packages everything an AI agent needs to work with the Codex API:- Operation map — Which query, mutation, or subscription to use for each task (pricing, charting, token discovery, events, wallets, holders, launchpads)
- Auth patterns — API key, bearer token, and MPP payment flows with header formats
- Query templates — Ready-to-use GraphQL for common operations like
filterTokens,getTokenPrices,getBars, and WebSocket subscriptions - Endpoint playbook — Decision guide for choosing the right operation based on intent
- Session preflight — Required
getNetworkscall to validate network IDs before making requests
Setup
- Claude Code
- OpenAI Agents
- Other agents
Clone the skill into your project:Claude Code will automatically pick up skills from the
.claude/skills/ directory.How it works
Once installed, your AI agent uses the skill to:- Pick the right operation — Maps intent (e.g. “get trending tokens”) to the correct GraphQL query (
filterTokens) - Set the correct auth — Chooses between API key headers and MPP payment flow based on context
- Generate valid GraphQL — Uses templates to produce runnable queries with proper variables and selection sets
- Follow best practices — Validates network IDs first, keeps selection sets minimal, batches subscriptions
Example
Ask your agent:“Get the top 10 trending tokens on Solana with prices and volume.”The skill guides it to generate: