AI-assisted development is the default way to build on Fireblocks. This page explains the tools available and when to use each.
Building on Fireblocks is designed to work with AI agents. Start by connecting your agent to the docs — the Documentation MCP is the one step that applies regardless of how you build. From there, use whichever combination of SDKs, CLI, and platform tools fits your workflow.
Get Started
The quickstart walks you through connecting your agent to the docs, setting up API authentication, and making your first Fireblocks API call — with SDK and CLI paths both covered.
What it is: An MCP server that gives your coding agent (Cursor, Claude Code, Codex, or any MCP-compatible tool) real-time access to Fireblocks developer documentation. Your agent can search, read, and cross-reference docs without leaving your IDE.When to use it: Always. Install this before anything else. It keeps your agent grounded in canonical, up-to-date documentation rather than training-data guesses — especially important for Fireblocks, where API behavior, authentication, and object models have specific details that general LLM training may not reflect accurately.How to use it:
Claude Code
Cursor
Codex
Other
claude mcp add --transport http fireblocks-docs https://developers.fireblocks.com/mcp
Point any MCP-compatible client at the Fireblocks Documentation MCP endpoint:https://developers.fireblocks.com/mcp
Once installed, your agent can answer questions like:
What fields does a vault account have?
How does Fireblocks JWT signing work?
What webhooks fire when a transaction is confirmed?
Without an MCP client: Every page on this site has a Copy Page button — use it to paste a page directly into Claude, ChatGPT, or any other LLM. You can also open the page URL in Claude’s Projects or any tool with web access. The MCP handles search and freshness automatically, but copying a specific page works well when you need focused context on a single topic.
Fireblocks publishes documentation in the llms.txt format for tools and workflows that expect a static, crawlable text bundle:
llms.txt — a compact index and entry point to the docs.
llms-full.txt — a broader aggregate of documentation content for setups that load one large file.
Use these when you cannot run an MCP client, need an offline or snapshot ingest for a custom index, or are wiring a tool that only understands llms.txt URLs. For everyday coding with an AI agent, the Fireblocks Documentation MCP (above) is the best option: it searches and retrieves the right pages on demand, reflects the current site, and avoids pulling huge static exports into context. Treat llms.txt / llms-full.txt as a fallback or supplement, not a replacement for the docs MCP.
What it is: An agent-first command-line tool that exposes every Fireblocks API operation as a typed, JWT-signed command. The help-index command returns a compact JSON index of all commands sized to fit in an LLM context window.When to use it: When you want to explore your workspace, prototype API calls, run scripts or CI jobs, or let your agent propose exact commands you can review before running. The CLI is the fastest path from “I want to do X” to a working, reviewable command — use it alongside an SDK in production, or on its own for operator and agent workflows.How to use it:
npm
Homebrew (macOS)
Standalone / other
npm install -g @fireblocks/fireblocks-cli
brew tap fireblocks/fireblocks-clibrew install fireblocks-cli
fireblocks help-index — compact JSON command catalogue for LLM context
--dry-run — preview a request before it executes
--no-confirm — skip interactive prompts in automated pipelines
Structured JSON errors on stderr with distinct exit codes
For any operation that creates or modifies data, ask your agent to show the fireblocks command first and use --dry-run where available. Review before running.
What it is: An MCP server that connects your AI tools (Claude, ChatGPT, Cursor) directly to your Fireblocks workspace data. Where the Documentation MCP gives your agent access to docs, AI Link gives it access to your live Fireblocks environment — balances, transactions, addresses, and workspace state.When to use it: When you want to query live workspace data in natural language, build dashboards, or run operational queries without writing API code. AI Link is oriented towards operations and treasury workflows rather than code generation.Two deployment modes:
Mode
Capabilities
Deployment
Best for
Remote MCP
Read-only
Fireblocks-hosted
Connecting to ChatGPT, Claude, or other external AI platforms
Local MCP
Read + write (transactions)
Self-hosted, open-source
Custom workflows with full data control
Example queries AI Link enables:
What percentage of my holdings are in stablecoins?
Summarize fees paid across all vaults this month.
What is the current balance of vault account 42?
Available via Fireblocks Labs for early access. The Local MCP server is open-source on GitHub.
What it is: A native AI assistant built directly into the Fireblocks Console — no external integrations required. Genie answers treasury and operations questions in real time using your workspace data, and can explain complex DeFi contract calls in plain language via the AI Transaction Summary feature.When to use it: When you’re in the Console and want fast answers about your workspace state, holdings, or a specific transaction — without switching context to another tool. Genie is not a developer tool; it’s built for treasury, finance, and operations teams who work in the Console.Capabilities:
Answer natural-language questions about balances, fees, and holdings
AI Transaction Summary: explains complex smart contract calls in human-readable terms
Respects your workspace policies, approval quorums, and user roles
The Documentation MCP is the one tool that applies to every workflow — install it first. After that, use the SDK, CLI, or both depending on what you’re building. Add AI Link when you want natural-language access to live workspace data from your own AI tools. Genie is available in the Console for non-developer users on your team.
