CAPP MCP ServerConnect Claude Desktop to Casper — grant, verify, and revoke AI agent permissions through natural language.
Read the spec →
Model Context Protocol

Claude can now manage
Casper agent permissions

The CAPP MCP Server exposes grant, verify, revoke, and audit as native MCP tools. Connect it to Claude Desktop — then grant AI agents bounded Casper authority through plain English. No blockchain UX. No manual deploys.

npm install -g capp-mcp-serverSetup guide →

5 MCP Tools

capp_grant

Grant an AI agent bounded signing authority using add_associated_key

agentPublicKeymaxTransferMoteskeyWeightexpiresInBlocks
capp_verify

Verify a Casper deploy was signed by a CAPP-authorized agent

deployHash
capp_revoke

Revoke agent authority — takes effect at next Casper block

agentPublicKey
capp_audit

List all agent permission entries — active, revoked, and expired

accountHashstatusFilter
capp_explain

Explain CAPP concepts: key-weight model, security, comparisons

topic

Claude + CAPP in action

Claude Desktop — CAPP connected
Grant my research agent 017d96b9e361... permission to transfer up to 100 CSPR on Casper testnet.
Tool call — capp_grant
{
  "agentPublicKey": "017d96b9e361...",
  "maxTransferMotes": "100000000000",
  "actionTypes": ["transfer"],
  "keyWeight": 1
}
Done. Agent granted with key weight 1.

Deploy hash: capp-grant-1a2b3c4d

Your account's deploy threshold is 3, so the agent cannot submit arbitrary deploys unilaterally — it's physically incapable of exceeding its mandate, not just contractually restricted.

Broadcast the deploy hash via CSPR.click to finalize on-chain.
What agents are currently authorized on my account?
Tool call — capp_audit
{ "statusFilter": "active" }
1 active agent permission found:

• 017d96b9e361... — transfer up to 100 CSPR
 Granted: just now · Deploy: capp-grant-1a2b3c4d · Status: active
Revoke it.
Tool call — capp_revoke
{ "agentPublicKey": "017d96b9e361..." }
Revocation prepared. Deploy hash: capp-revoke-9f8e7d6c

Broadcast this deploy to finalize. The agent will lose signing authority at the next Casper block after inclusion.

Setup — Claude Desktop

1. Install the server

npm install -g capp-mcp-server

2. Add to Claude Desktop config

Edit ~/Library/Application Support/Claude/claude_desktop_config.json

{
  "mcpServers": {
    "capp": {
      "command": "npx",
      "args": ["capp-mcp-server"],
      "env": {
        "CAPP_NETWORK": "testnet",
        "CAPP_SIGNER": "<your-account-public-key>"
      }
    }
  }
}

3. Start talking to Casper

“Grant my agent 017d96b9... permission to transfer up to 50 CSPR”
“What agents are currently authorized on my account?”
“Verify that deploy abc123... was signed by an authorized agent”
“Revoke access for agent 017d96b9...”
“Explain how CAPP's key weight model works”

Why this is a first

First MCP-native blockchain permission layer

No other chain has an MCP server for AI agent authority management. CAPP makes Casper the native home of Claude-powered on-chain agents.

Natural language → cryptographic enforcement

"Grant 100 CSPR" becomes an add_associated_key deploy enforced by every Casper validator. The gap between human intent and blockchain execution disappears.

Works with any MCP-compatible AI

Claude Desktop, Cursor, Continue, or any tool implementing the Model Context Protocol can call CAPP tools — the standard outlives any single AI product.

View source on GitHub →