Safeheron Remote MCP
This tutorial walks through connecting, authorizing, and using Safeheron Remote MCP to interact with your Safeheron team via AI Agents.
1. Product Overview
Safeheron Remote MCP enables AI Agents (such as Claude or Cursor) to connect directly to your Safeheron workspace. Instead of navigating complex interfaces, you simply describe what you need in natural language.
Key Benefits:
Natural Language Interaction: Replace complex UI navigation with everyday language, drastically lowering the learning curve.
Zero Installation: Just paste a URL and complete a browser-based authorization — no software to install.
Managed Service: Safeheron handles all infrastructure, updates, and maintenance.
What You Can Do:
Query all wallet accounts and balances
Retrieve recent transaction records
View aggregated asset balances across wallets
Inspect approval policies and audit nodes
Search whitelist addresses
2. Prerequisites
Before getting started, make sure the following are in place:
Safeheron Workspace
An active, configured Safeheron account
Remote MCP Feature
Not enabled by default — You can enable this feature via Web Console at Tools > Safeheron Lab
AI Client
Claude Desktop / Claude Web (Pro/Max) or Cursor
Browser
Required for the OAuth authorization flow
Note: Once Remote MCP is enabled, every user in the team can connect an AI Agent. Confirm your team's security policy before enabling.
3. Connection & Authorization
3.1 Claude Desktop / Web
Step 1: Copy the MCP Server URL
Step 2: Configure Claude Connector
Open Claude Desktop or Web, navigate to Settings → Connectors → Go to customize, click Add custom Connector, and enter a name (e.g., Safeheron) along with the URL above.


Step 3: Connect & Authorize
Click the Connect button next to the newly added MCP Server. Your browser will open the Safeheron OAuth authorization page. Confirm authorization to establish the connection.


Step 4: Start Chatting
Type the following in Claude's chat:
If everything is set up correctly, Claude will call Safeheron MCP tools and return your asset information.

3.2 Cursor
Step 1: Copy JSON Configuration
Step 2: Configure Cursor MCP Server
Open Cursor → Cursor Settings → Tools & MCP, click Add Custom MCP Server, and paste the JSON above.


Step 3: Connect & Authorize
Click the Connect button next to the MCP Server and complete browser authorization.



Step 4: Start Chatting
Type in Cursor's AI chat:
4. Core Features in Action
Safeheron Remote MCP provides 12 core tools organized into 6 use-case scenarios below.
4.1 Asset Overview
Scenario: Quickly understand the overall asset status of your workspace.
Example Prompt:
Tool Used: getWorkspaceAssets
Expected Result: The AI returns the workspace's total USD-denominated asset value.
Advanced Prompt:
Tool Used: queryMultiAccountBalanceStat
Expected Result: Aggregated BTC and ETH balances summed across all wallets.

4.2 Wallet Management
Scenario: Browse wallets, search by name, and inspect coin details.
Example Prompt 1: List All Wallets
Tool Used: listWallets
Expected Result: A list of wallet IDs, names, and USD balances.
Example Prompt 2: Search by Name
Tool Used: listWallets (with namePrefix parameter)
Example Prompt 3: Wallet Coin Details
Tools Used: listWallets → getWalletCoins
Expected Result: The AI first locates the Treasury wallet ID, then queries all coin balances and addresses within it.



4.3 Transaction Search & Tracking
Scenario: Search transaction history and track specific transactions.
Example Prompt 1: Recent Transactions
Tool Used: searchTransactions
Example Prompt 2: Filtered Search
Tool Used: searchTransactions (with time and amount filters)
Example Prompt 3: Track a Specific Transaction
Tool Used: getTransaction
Transaction Status Reference:
0
SUBMITTED
Pending approval
1
CANCELLED
Cancelled
2
BROADCASTING
Broadcasting to blockchain
3
CONFIRMING
Awaiting block confirmations
4
COMPLETED
Transaction complete
5
FAILED
Transaction failed
7
REJECTED
Rejected by approvers
10
SIGNING
Being signed



4.4 Approval Policy
Scenario: Understand the current transaction approval rules and audit node configuration.
Example Prompt 1: View Policy
Tool Used: queryActivePolicy
Expected Result: The AI interprets and explains the active policy in natural language, covering initiator restrictions, source wallet constraints, destination rules, asset types, and amount thresholds.
Example Prompt 2: List Audit Nodes
Tool Used: listAuditNodes
Expected Result: Returns each audit node's name, status, approver list, and associated rules.


4.5 Whitelist Management
Scenario: Search pre-approved whitelist addresses.
Example Prompt:
Tool Used: searchWhitelist
Tool Used: searchWhitelist (without search parameters)


5. Best Practices
5.1 Prompting Tips
Use specific, clear natural language to describe your needs:
Check assets
Show my total workspace assets in USD
Recent transactions
List all completed transactions over 500 USD from the past 7 days
Transfer
Send 0.5 BTC from the Treasury wallet to whitelist address "Binance Hot Wallet" with medium fee level
5.2 Multi-Step Task Orchestration
You can ask the AI to handle complex, multi-step tasks:
The AI will automatically orchestrate multiple tool calls and produce a consolidated report.
5.3 Policy Auditing
Periodically review your approval policy configuration via AI:
5.4 Transaction Monitoring
Use AI for routine transaction monitoring:
6. Security Notes
Read-Only by Default: Remote MCP is configured for read-only operations. AI Agents cannot initiate or sign transactions unless write access is explicitly enabled.
Review AI Output: AI models may produce unintended results. Always carefully verify AI-returned data and suggested actions, especially amounts and addresses.
Authorization Scope: Connecting an AI Agent requires OAuth authorization, after which the AI can query data on your behalf. Only connect trusted AI clients.
Workspace-Wide Access: Once enabled, all users in the workspace can connect AI Agents. Ensure your team is aware of the security guidelines.
7. FAQ
Q1: How do I enable Remote MCP?
Contact Safeheron support at support@safeheron.com to request access.
Q2: Which AI clients are supported?
Currently supported: Claude Desktop, Claude Web (Pro/Max users), and Cursor. Any MCP-compatible AI client should theoretically work.
Q3: Can the AI directly transfer my assets?
No. The default mode is read-only. Even with write access enabled, your explicit chat confirmation plus Safeheron's approval policy provide a double safeguard.
Q4: How do I disconnect?
Remove the corresponding Connector in your AI client's Settings → Connectors.
Q5: Which blockchains and coins are supported?
Ask the AI for the latest list:
Q6: What if authorization fails?
Verify that: (1) Remote MCP is enabled for your workspace; (2) your Safeheron account is active; (3) your browser isn't blocking the OAuth popup. If issues persist, contact support@safeheron.com.
Last updated

