Reference

Connect a client

TuskPoint speaks the Model Context Protocol over stdio, so any MCP-capable client can use all eight tools. Pick your client below — the config is the same shape everywhere; only the file location changes.

One config, four clients

Every client uses the same { "mcpServers": { "tuskpoint": { … } } } block. Set cwd to the absolute path of your cloned repo so mcp_server/server.py resolves. Or just call the tuskpoint_info tool and let the agent emit the right snippet for you.

Claude Desktop

Edit claude_desktop_config.json (Settings → Developer → Edit Config) and add the server, then restart Claude Desktop.

claude_desktop_config.json

{
  "mcpServers": {
    "tuskpoint": {
      "command": "python",
      "args": ["mcp_server/server.py"],
      "cwd": "/absolute/path/to/tuskpoint",
      "env": {
        "WALRUS_AGGREGATOR_URL": "https://aggregator.walrus-mainnet.walrus.space",
        "WALRUS_PUBLISHER_URL": "https://walrus-mainnet-publisher-1.staketab.org:443"
      }
    }
  }
}

Claude Code

terminal

claude mcp add tuskpoint -- python mcp_server/server.py

Then confirm it is wired up:

terminal

claude mcp list

Cursor

Create .cursor/mcp.json in your project (or the global ~/.cursor/mcp.json) with the same block.

.cursor/mcp.json

{
  "mcpServers": {
    "tuskpoint": {
      "command": "python",
      "args": ["mcp_server/server.py"],
      "cwd": "/absolute/path/to/tuskpoint",
      "env": {
        "WALRUS_AGGREGATOR_URL": "https://aggregator.walrus-mainnet.walrus.space",
        "WALRUS_PUBLISHER_URL": "https://walrus-mainnet-publisher-1.staketab.org:443"
      }
    }
  }
}

Windsurf

Add the block to ~/.codeium/windsurf/mcp_config.json and reload the MCP servers from the Cascade panel.

~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "tuskpoint": {
      "command": "python",
      "args": ["mcp_server/server.py"],
      "cwd": "/absolute/path/to/tuskpoint",
      "env": {
        "WALRUS_AGGREGATOR_URL": "https://aggregator.walrus-mainnet.walrus.space",
        "WALRUS_PUBLISHER_URL": "https://walrus-mainnet-publisher-1.staketab.org:443"
      }
    }
  }
}

Any other MCP client

A ready-to-use .mcp.json ships in the repo root. Point your client at it, or copy the block above. The transport is stdio and the command is always python mcp_server/server.py.

Optional: enable semantic search

To turn on checkpoint_search, add your MemWal credentials to the env block (or your .env):

env

MEMWAL_PRIVATE_KEY=your_ed25519_delegate_key_hex
MEMWAL_ACCOUNT_ID=your_walrus_memory_account_object_id

Writes need a publisher

Reads work out of the box. To checkpoint_save or checkpoint_fork on mainnet you need a publisher with a funded key — see Walrus mainnet for the options.