MCP server
Install @fabricbloc/mcp. Any MCP-aware agent gets wallet, token, vault, and balance tools through a local, secret-backed key.
Coming soon · Agentic finance
One agent for onchain rails.
The FabricBloc agent reaches your wallets, tokens, and vaults over MCP, the SDK, or an assistant like Claude or Codex. One scoped key, one audit log, one surface — coming soon.
npx @fabricbloc/mcp{
"fabricbloc": {
"command": "npx",
"args": ["@fabricbloc/mcp"]
}
}// Issue a scoped key — agents only get what they need
const key = await fb.keys.createScoped({
permissions: [
"wallets:read",
"tokens:transfer",
"vaults:deposit",
],
expiresIn: "24h",
maxSpend: { asset: "USDC", amount: "1000" },
});One capability, many reach paths
Works where your agent already runs.
Claude plugin
One-click Claude.ai integration. Claude can read balances, execute transfers, and manage vault positions.
Codex plugin
OpenAI Codex and GPT function-calling support. Pass a scoped key; agents operate within defined limits.
Agent SDK
TypeScript and Python SDKs with agent-first patterns — retry, idempotency, and spend-cap enforcement built in.
MCP server · available now
Install the MCP server.
The Model Context Protocol server is the one piece you can use today. One config entry exposes FabricBloc wallet, token, vault, and balance tools to any MCP-aware host. The scoped key stays in your local environment.
Install
npx @fabricbloc/mcp Configure
~/.claude/mcp.json ~/.claude/mcp.json
{
"fabricbloc": {
"command": "npx",
"args": ["@fabricbloc/mcp"]
}
}Reach paths
Same agent, your runtime.
Reach the same FabricBloc tools from an MCP host, the Codex CLI, GPT function-calling, or your own SDK-built host — all bound by the same scoped keys, spend caps, and audit log.
MCP host
Claude Desktop, Cursor, or any MCP-aware runtime — point it at the server.
npx @fabricbloc/mcp Codex CLI
Pass the MCP server as a tool endpoint in your Codex task.
codex --mcp-server npx:@fabricbloc/mcp OpenAI tools
Drop FabricBloc tools into GPT function-calling with a scoped key.
import { fabricblocTools } from '@fabricbloc/openai-tools'; SDK
Build your own agent host on the TypeScript or Python SDK.
import { FabricBloc } from '@fabricbloc/sdk'; Use cases
What agents do with FabricBloc.
Autonomous treasury
Agent rebalances vault positions on a schedule. Deposits idle USDC when yield exceeds threshold. No manual intervention.
Agent-to-agent payments
USDC settlement between agents in a multi-agent system. Scoped caps prevent overspend. Every transfer is auditable.
AI customer service
Refund, reissue, or top-up customer wallets from a chat agent. Agent acts on your behalf with bounded permissions.
Onchain research
Agents read balances and transaction history to summarize account activity, flag anomalies, or generate reports.
Security model
Agents get only what they need.
Scoped API keys define exactly which operations an agent can perform and how much it can spend. The FabricBloc API enforces these limits server-side — agents cannot escalate permissions.
- Granular permissions:
wallets:read,tokens:transfer,vaults:deposit - Per-asset spend caps with hard limits
- Key expiry — automatic revocation after TTL
- Full audit log for every action taken by a key
// Issue a scoped key — agents only get what they need
const key = await fb.keys.createScoped({
permissions: [
"wallets:read",
"tokens:transfer",
"vaults:deposit",
],
expiresIn: "24h",
maxSpend: { asset: "USDC", amount: "1000" },
});FAQ · structured for AI search
Common questions about agents.
What is the FabricBloc MCP server?
An MCP (Model Context Protocol) server that exposes FabricBloc wallet, token, vault, and balance operations as tools any MCP-aware AI agent can call.
How do I prevent an agent from spending too much?
Issue a scoped API key with explicit permission limits (wallets:read, tokens:transfer) and an optional maxSpend cap per asset. The API enforces these server-side.
Which AI frameworks are supported?
Claude (claude.ai and API), OpenAI GPT function-calling, and any MCP-compatible agent runtime. The Agent SDK provides idiomatic wrappers for TypeScript and Python.
Is agent activity auditable?
Yes. Every wallet operation, transfer, and vault action is logged with the key ID and timestamp. You can query the audit log via the API or console.
Do agents need their own wallets?
Not necessarily. Agents can operate on behalf of user wallets with scoped permissions, or you can provision a dedicated agent wallet for treasury operations.
Give your agent a wallet.
The FabricBloc agent is coming soon. The MCP server is available now.