Index of every page on this site: llms.txt
All skills
base avatar

/base-mcp

@ef8bcd5 official
by basebase/skills120 stars
130
Run it yourself

Base MCP — gives your AI assistant access to a Base Account via the Base MCP server (wallet-mcp.coinbase.com). Wallet, portfolio, sending, swapping, signing, x402 payments, batched contract calls, and transaction history across supported chains.

Use this Skill: https://skilld.dev/gh/base/skills/base-mcp

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referencesapproval-mode.md

≈750 tokens on demand. Your agent reads this file only when SKILL.md points to it.

Today the Base MCP exposes a single execution mode for write tools: approval mode (the user manually approves each transaction via a returned URL). Other modes may be added later. Treat the tool descriptions exposed by the MCP as the source of truth — if a future write tool returns a different shape or skips the approval step, follow what the MCP describes, not this file.

Approval Mode

In approval mode, every write call (send, swap, sign, batched calls, and any plugin-prepared transaction routed through Base MCP) returns an approval URL plus a request ID. The user opens the URL, approves the action in Base Account, and then the agent polls the request ID for completion.

Flow

  1. Call the write tool. The response includes:

    • an approval URL (the field name is on the MCP response — typically approvalUrl)
    • a request ID (typically requestId)

  2. Show the user the link. Present it as "Approve Transaction" (or similar neutral language). Refer to the approval destination as Base Account, not as the raw URL hostname or an implementation-specific provider. Just give the user the link to click.

  3. In CLI harnesses, also open the link automatically. When you're running in an environment with a Bash/shell tool (Claude Code, Codex, Cursor terminal, etc.), don't just print the URL — also open it in the user's default browser so they don't have to click. Always print the link too as a fallback, then run the platform-appropriate open command:

    • macOS: open "<url>"
    • Linux: xdg-open "<url>" (fall back to wslview under WSL)
    • Windows: start "" "<url>" (or via PowerShell: Start-Process "<url>")

    Skip this step on chat-only surfaces (ChatGPT, Claude.ai) — they don't have a shell, so just show the link.

  4. Wait for the user to confirm they approved. Don't poll in a tight loop while they're still acting.

  5. Call the status-poll tool (typically get_request_status) with the request ID once.

  6. Only report success when the status tool confirms completion.

Common mistakes

  • Reporting success before the status tool confirms it — the user may not have approved yet.
  • Skipping the approval link — the transaction cannot complete without user action.
  • Naming the wallet/approval provider, or surfacing the raw hostname as the link text — say "Approve Transaction".
  • Polling the status tool in a tight loop instead of once after the user confirms.
  • Forgetting to also auto-open the link in CLI harnesses where a shell is available — printing alone makes the user copy-paste unnecessarily.
  • Trying to auto-open in chat-only harnesses where no shell exists — that just produces an error.

Source: SKILL.md on GitHub

Skill behaviors

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

Third-party checks

1 alert14d3 checks · Risk CRITICAL
  • Gen Agent Trust Hub14d

    Security analysis identified a critical risk due to automated scanner detections of malicious files and URLs within the Flaunch plugin. The skill also contains hardcoded authentication credentials across multiple plugins and relies on downloading and executing various third-party packages at runtime. Furthermore, certain plugins handle highly sensitive personal information, such as credit card details and email contents, which increases the potential impact of an exploit.

  • Socket14d

    1 alert: gptSecurity

  • Snyk14d

    Risk: MEDIUM · 2 issues

Provenance

Signed by skilld at ef8bcd5. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 2 weeks ago.

Activeupdated 2 weeks ago

Capability

title
Base MCP Skill
version
0.1.0

Topics

  • MCP
  • base
  • blockchain
  • wallet
  • defi
  • ethereum
  • swaps
  • tokens
  • agent-tools

README badge

README badge for base/skills/base-mcp

What it does

Gives your AI assistant access to a Base Account wallet via the Base MCP server, enabling balance checks, token sends and swaps, message signing, x402 payments, batched contract calls, and transaction history. Includes optional plugins for Morpho, Moonwell, Uniswap, Avantis perpetuals, Virtuals agents, Aerodrome, and Bankr token launches.

Generated from the current SKILL.md.

Frequently asked

What do I need to install to use this skill?
You need the Base MCP server installed and connected. If Base MCP tools aren't available, visit https://docs.base.org/ai-agents/quickstart for installation steps specific to your platform.
What chains does Base MCP support?
The skill works across supported chains for wallet, portfolio, and transaction history. Individual plugins (Morpho, Moonwell, Uniswap, Avantis, Aerodrome, Bankr) are Base-only unless noted otherwise — Moonwell also supports Optimism.
Can I use this skill to sign transactions?
Yes. Base MCP supports signing messages and contract calls. Write operations that return an approval URL follow a specific approval flow — the skill will guide you through it.
What happens if I use a plugin that supports lending or leverage?
Plugins like Morpho, Moonwell, and Avantis can expose you to liquidation risk. The skill surfaces health factor and position data before risky actions, but you must review all details before approving any transaction.
Do I need to provide my wallet address at the start?
No. Wallet address and balance are only fetched when you ask for them or when a pending operation actually needs the address — like a token send or balance lookup.

Generated from the current SKILL.md. These answers refresh after source changes.