In a major leap forward for the intersection of artificial intelligence and Web3, Coinbase’s Layer-2 network, Base, has officially launched Base MCP (Model Context Protocol). The newly released tool establishes a secure, standardised bridge connecting users’ Base Accounts directly to popular AI interfaces and agentic environments, allowing users to execute complex onchain transactions using simple conversational prompts.
At launch, Base MCP supports some of the industry’s leading AI clients, including Claude (Web, Desktop, and Code), ChatGPT, Codex, and Cursor, allowing users to manage portfolios, execute swaps, and interact with decentralised finance (DeFi) protocols directly from their active chat windows.
Bridging the Gap Between Natural Language and Web3
The launch represents a paradigm shift in how users interact with blockchain technology. Traditionally, onchain actions have required manual navigation through multi-step decentralised application (dApp) frontends, wallet connections, and complex transaction approvals.
Base MCP simplifies this user experience by turning AI agents into active facilitators. By linking a Base Account to an AI interface, users can simply type prompts such as:
- “Check my current portfolio balances across Base.”
- “Swap 0.5 ETH for USDC.”
- “Explore the latest token launches and buy £50 worth of the trending agent token.”
Rather than acting on its own isolated intelligence, the AI agent is granted access to the user’s active wallet state via a secure API connection, translating colloquial requests into precise, executable transactions.
Ecosystem Integration: Skill Plugins from Day One
To ensure utility from its release, Base MCP has debuted with a suite of “Skill Plugins” developed in collaboration with prominent protocols across the Base ecosystem. These plugins allow AI agents to navigate beyond simple wallet transfers and tap into complex financial structures.
At launch, the ecosystem integrations include:
- Uniswap & Aerodrome: Allowing agents to explore liquidity pools, review LP positions, manage liquidity, and execute asset swaps.
- Morpho & Moonwell: Allowing users to find yield opportunities, supply assets to lending vaults, and borrow against collateral.
- Avantis: Facilitating the exploration and management of onchain perpetual trading positions.
- Bankr & Virtuals Protocol: Designed for the agentic economy, allowing users to discover recently launched tokens and purchase AI-driven agent assets natively from chat.
Developers looking to integrate their own protocols can write a concise markdown specification that teaches the AI assistant how to pull unsigned transaction details from their API, integrating them seamlessly into the Base MCP flow.
Under the Hood: Security and User Control Stand Firm
One of the largest hurdles for AI-driven onchain systems has been security. Allowing automated agents to interact with financial infrastructure raises obvious concerns regarding unauthorised drainings, prompt injection attacks, and key management vulnerabilities.
Base has structured the MCP framework with a strict “User-in-the-Loop” philosophy to mitigate these risks:
1. Zero Key Access
The MCP server never has access to the user’s private keys. Instead, when an agent proposes an action, it constructs the transaction locally and submits it to a temporary queue.
2. Sandbox Review and Simulation
The agent provides the user with a secure transaction link. Clicking this link opens the user’s Base Account in a separate, isolated browser window. Here, the user is presented with a simulation showing the exact asset changes that will occur. No funds can move without an explicit, manual confirmation from the user.
3. Phishing Mitigation
Because transactions are constructed locally by the user’s trusted AI environment rather than retrieved from a external website, Base MCP bypasses traditional web-vulnerabilities such as domain hijacking and DNS spoofing.
4. Enterprise-Grade Authentication
The authentication pipeline is built on OAuth 2.1, the global industry standard powering “Sign in with Google” systems, adapted natively to support onchain cryptographic web accounts.
Technical Mechanics: Stored Requests
The transaction flow relies on a pre-existing Base Account primitive known as stored requests—a system originally designed to streamline retail checkouts through Shopify Base Pay.
When an AI agent triggers an action, the wallet’s JSON-RPC request is securely stored on the server, generating a unique request ID. The MCP server wraps this ID into a dynamic URL. When the user opens the link, the Base Account pulls the pending request, displays the clear-text transaction variables to the user, and signs the payload upon confirmation.
The agent polls for this signature status and immediately resumes its automated task flow once the block has been confirmed onchain.
What’s Next for the Onchain Agent Era?
The release of Base MCP signals Coinbase’s aggressive push to position Base as the default network for the burgeoning “agentic economy.” By making AI agents fully contextually aware of a user’s onchain history, assets, and preferences, Base has laid the groundwork for highly sophisticated autonomous assistants.
In its official release statement, the Base development team noted:
“We will keep improving Base MCP with more skill plugins, better prompt flows, and deeper support for the ways people use Base from their favourite AI interfaces.”
As AI models continue to evolve from simple text-generation engines to active onchain participants, tools like Base MCP will likely prove critical in shifting blockchain technology from a niche developer sandbox to an intuitive, conversational consumer utility.
Developers and enthusiasts can begin configuring Base MCP today by visiting the official Base documentation quickstart.