EmblemAI Agent Skills is a skill collection officially published by EmblemAI, intended for AI agents built with EmblemAI. EmblemAI is open-source crypto infrastructure for end users and AI agents, providing over 200 tools across 7 blockchains (Solana, Ethereum, Base, BSC, Polygon, Hedera, Bitcoin) for token swaps, conditional orders, DeFi, NFTs, and cross-chain bridges, with support for the x402 payment channel as well as A2A and MCP protocols.
Key capabilities
- For AI agents and developers: 200+ tools covering 7 blockchains, including swaps, conditional orders, DeFi positions, NFT operations, cross-chain bridges, and market intelligence; x402 facilitator and pay-per-call channels, enabling agent-to-agent commerce without account setup; native MCP (Model Context Protocol) server, suitable for MCP clients such as Claude Code, GitHub Copilot, and Gemini CLI; A2A (Agent-to-Agent) protocol support, enabling direct agent interoperability; deterministic wallet—one password generates one persistent agent identity across all chains, and every operation that modifies the wallet requires explicit approval.
- For end users and wallet-native applications: user management for wallet-native applications, creating website users who also have a full crypto wallet with a single integration; flexible login, allowing users to sign in through multiple crypto wallets, email/password, or social login; through Emblem's agent wallet CLI, create or restore a wallet-capable identity with a single command; the same Emblem session can simultaneously support authentication, wallet access, transaction signing, and AI-driven crypto workflows.
How to use
- Install as an MCP server: The above skills are authoring guidance. To have an agent actually call EmblemAI tools at runtime, install the hosted MCP server in an MCP-compatible client.
- Claude Code (OAuth, no API key required): Claude Code guides you through the hosted OAuth flow in the browser, with no need to paste a key.
- Claude Code (API key, for headless/CI): Provides the corresponding installation method.
- GitHub Copilot CLI / .mcp.json: Provides the corresponding configuration.
- Full installation matrix (Claude Desktop bridge, Cursor, Windsurf, Gemini CLI):https://emblemvault.ai/docs/mcp
Skill Structure Each skill follows the Agent Skills specification. See CONTRIBUTING.md for guidelines on adding new skills.
Registry Compatibility Publishable SKILL.md files in the repository are intentionally kept within the public Agent Skills frontmatter scope defined by agentskills.io/specification: name, description, and the optional license, compatibility, metadata, and allowed-tools fields. This makes these skills portable across agentskills.io / agentskills.to compatible tools without relying on vendor-proprietary top-level fields.
The public documentation describes the repository-based sharing and installation workflow, not the manual "Add Skill" form documented in the Web UI. In practice, keep skills in git, validate them with the official CLI, and share or install them from the repository.
The upstream agentskills validate command validates one skill directory at a time, not the top-level ./skills folder. This repository's validate-all.sh wrapper script iterates over skills/*/, syncs the shared EmblemAI reference sources into the skill-local copies, and then runs repository-specific markdown/link checks.
The legacy --strict flag is still accepted by the wrapper script for backward compatibility, but it no longer changes behavior, because the official validator conforms to the specification by default.
Pull requests install the official skills-ref validator in CI and run the repository wrapper script to keep public skill frontmatter conformant with the specification. CI also fails if the sync step rewrites generated files under skills/, preventing uncommitted shared reference updates from being missed.
Some reference documents are intentionally duplicated in publishable skills. These duplicated files should not be edited in place under skills/.../references/: edit the canonical source under shared/, run bash validate-all.sh or bash validate-skill.sh ..., and let the sync step regenerate the skill-local copies.
Current shared sources: shared/emblem-ai-prompt-examples.md, shared/emblem-ai-prompt-examples/, shared/emblem-ai-react-references/. Current generated skill-local copies: prompt examples copied to emblem-ai, emblem-ai-react, emblem-ai-agent-wallet, and emblem-ai-prompt-examples; React references copied to emblem-ai and emblem-ai-react. The sync entry point is utils/sync-emblem-ai-shared-references.sh. If a reference file is duplicated across multiple skills, it should be moved into shared/ and synced back to each individual skill, rather than maintaining separate manual copies. Validation also runs automatically on every PR via GitHub Actions.
When to Use
- Development scenarios that need to connect multi-chain crypto operation capabilities to AI agents.
- Scenarios for building wallet-native applications that need user management, authentication, and wallet integration.
- Scenarios that need agent-to-agent payments (x402) or A2A interoperability.
- Agent environments compatible with Agent Skills or MCP, such as Claude Code, Cursor, Codex, GitHub Copilot, and Gemini CLI.
Limitations
- Every operation that modifies the wallet requires explicit approval.
- Skills follow the Agent Skills specification, with frontmatter limited to public fields, and may be incompatible with tools that rely on vendor-proprietary fields.
- The upstream validation command validates only one skill directory at a time, so the repository wrapper script is needed for batch validation.
- Some reference documents are duplicated across multiple skills and must be maintained via sync from the shared/ source.
EmblemAI provides 200+ autonomous trading tools across 7 blockchains, is open source, and can be used as an npm CLI, MCP server, and A2A-compatible agent.
Related links:
- Website: emblemvault.ai
- Blog: emblemvault.ai/blog
- Docs: emblemvault.ai/docs (specification) · emblemvault.dev (interactive)
- NPM: @emblemvault/agentwallet
