Skills and MCP Usage

This page explains how the local skills/ library and mantle-mcp are intended to work together. The repository keeps the skills/ submodule path stable, but the actual skill folders come from mantle-xyz/mantle-skills under skills/skills/.

Operating Model

Skills provide workflow and decision guidance.
mantle-mcp provides deterministic data access and structured outputs.

A reliable agent loop is:

Choose the right skill for the task.
Use MCP discovery interfaces (listTools, listResources, listPrompts).
Execute tool calls with validated inputs.
Ground the final answer in tool/resource/prompt outputs.

How To Use Skills

All project skills live in skills/skills/<skill-name>/SKILL.md with references and agent manifests. If the checkout is missing after clone, run npm run skills:init to initialize the mantle-skills submodule.

Recommended usage flow:

Map user intent to one primary skill (or a small sequence).
Read that skill’s SKILL.md before tool execution.
Follow its required checklist and safety constraints.
Use the MCP tool/resource/prompt surface to execute the workflow.

For cross-environment integration guidance, see External Agent Integration.

Current Skill Catalog

Skill	Primary Use	Typical MCP Touchpoints
`mantle-network-primer`	Explain Mantle architecture and network basics	`mantle_getChainInfo`, `mantle_getChainStatus`, `mantle://docs/network-basics`
`mantle-address-registry-navigator`	Resolve and validate canonical addresses	`mantle_resolveAddress`, `mantle_validateAddress`, `mantle://registry/contracts`
`mantle-portfolio-analyst`	Wallet balance/allowance portfolio review	`mantle_getBalance`, `mantle_getTokenBalances`, `mantle_getAllowances`, `mantle_getTokenPrices`
`mantle-data-indexer`	Subgraph + SQL data retrieval with quality controls	`mantle_querySubgraph`, `mantle_queryIndexerSql`
`mantle-defi-operator`	DeFi read workflows for swaps, pools, lending	`mantle_getSwapQuote`, `mantle_getPoolLiquidity`, `mantle_getLendingMarkets`
`mantle-readonly-debugger`	Diagnose read-path failures and tool errors	`mantle_checkRpcHealth`, `mantle_probeEndpoint`, error payload codes
`mantle-risk-evaluator`	Risk scoring and execution guardrails	`mantle://docs/risk-checklist`, `mantle_getAllowances`, `mantle_getTokenPrices`
`mantle-tx-simulator`	Pre-signing transaction simulation guidance	prompt/resource guidance + simulation-oriented tool sequencing
`mantle-smart-contract-deployer`	Deployment/verification checklist orchestration	chain/config resources + diagnostics checks before/after deploy

How To Use mantle-mcp in Practice

For most production workflows, start with this sequence:

Check network context: mantle_getChainInfo, mantle_getChainStatus
Resolve identifiers first: mantle_resolveAddress / mantle_resolveToken
Run domain tools (account, token, DeFi read, indexer)
Add diagnostics checks when any endpoint is unstable

This keeps the agent grounded in canonical Mantle metadata and reduces hallucinated addresses/endpoints.