Support Dynamic Workspace/Server Switching for Multi-Server Environments

[bkatkam123]

Summary

The p4 MCP server currently uses static P4PORT/P4CLIENT configuration set at startup, making it impossible to query multiple Perforce servers or switch workspaces dynamically. This significantly limits the tool's usefulness in enterprise environments with multiple Perforce servers (e.g., product lines on different servers) and forensic/investigation workflows that require analyzing data across different depots.

Environment

• P4 MCP Server Version: 2025.1.0
• MCP Client: Claude Code / Anthropic
• Infrastructure: Multi-server Perforce environment
  • Server 1: server1.example.com:1666 (Corporate tools - //corp_tools depot)
  • Server 2: server2.example.com:1666 (Product development - //product_a, //product_b depots)
• Use Case: Forensic analysis, cross-depot investigation, multi-project development

Current Behavior

The MCP server is configured with static environment variables at startup:

{
  "mcpServers": {
    "perforce-p4-mcp": {
      "command": "/path/to/p4-mcp-server",
      "env": {
        "P4PORT": "server1.example.com:1666",
        "P4CLIENT": "workspace1"
      }
    }
  }
}

Limitations:

1. ❌ All queries go to the same P4PORT server
2. ❌ All queries use the same P4CLIENT workspace
3. ❌ Cannot access depots on other Perforce servers
4. ❌ Cannot switch workspace context during session
5. ❌ Shell environment variables (export P4PORT=...) have no effect
6. ❌ Working directory changes (cd /workspace) are ignored

Expected/Desired Behavior

Support one or more of the following approaches to enable dynamic workspace/server switching:

Option 1: Per-Query Server/Workspace Override (Preferred)

Allow query tools to accept optional parameters to override the default connection:

{
  "action": "history",
  "file_path": "//product_a/main/lib/example_cell/schematic/sch.oa",
  "p4_config": {
    "P4PORT": "server2.example.com:1666",
    "P4CLIENT": "workspace2"
  }
}

Benefits:

• ✅ Flexible per-query configuration
• ✅ Single MCP server can access multiple P4 servers
• ✅ Backward compatible (optional parameter)
• ✅ Most powerful solution

Option 2: P4CONFIG Directory Support

Allow queries to specify a directory containing .p4config file:

{
  "action": "history",
  "file_path": "//product_a/main/lib/example_cell/schematic/sch.oa",
  "p4config_dir": "/workspace/product_a"
}

The MCP server would read .p4config from that directory to determine P4PORT/P4CLIENT.

Benefits:

• ✅ Leverages existing P4CONFIG mechanism
• ✅ Familiar to Perforce users
• ✅ Works with existing workspace configurations

Option 3: Workspace Switching Tool

Add a new tool to dynamically switch the active workspace:

{
  "tool": "set_workspace",
  "params": {
    "P4PORT": "server2.example.com:1666",
    "P4CLIENT": "workspace2"
  }
}

All subsequent queries would use the new configuration until changed again.

Benefits:

• ✅ Simpler than per-query override
• ✅ Efficient for multiple queries to same server
• ⚠️ Stateful (queries affect each other)

Option 4: Multiple Named Connections

Allow MCP configuration to define multiple named connections:

{
  "mcpServers": {
    "perforce-p4-mcp": {
      "command": "/path/to/p4-mcp-server",
      "connections": {
        "corporate": {
          "P4PORT": "server1.example.com:1666",
          "P4CLIENT": "workspace1"
        },
        "products": {
          "P4PORT": "server2.example.com:1666",
          "P4CLIENT": "workspace2"
        }
      },
      "default_connection": "corporate"
    }
  }
}

Queries could specify which connection to use:

{
  "action": "history",
  "connection": "products",
  "file_path": "//product_a/..."
}

Benefits:

• ✅ Pre-configured trusted connections
• ✅ Simple connection selection per query
• ✅ Secure (no arbitrary server access)

[MohitPerforce]

Hello @bkatkam123
Thanks for the detailed proposal ,this is very helpful and we understand the limitations you’re highlighting with the current static P4PORT/P4CLIENT model.

Before we evaluate solutions, could you confirm:
• Is your primary need forensic/multi-server investigation or day-to-day multi-project usage?
• Do you need concurrent access to multiple servers/workspaces, or is switching context sufficient?

This will help us assess the best approach and next steps.

Additionally, it may be useful to note that we already provide:

• Option 2: P4CONFIG Directory Support
  You can set the P4CONFIG environment variable in the MCP server configuration. Alternatively, if no explicit configuration is provided, the MCP server will automatically pick up the Perforce configuration from the current working directory. You can find these details here.

• Option 3: Workspace Switching Tool
  We provide a modify_workspaces tool that also supports switching workspaces. This capability is documented here:
  https://github.com/perforce/p4mcp-server?tab=readme-ov-file#modify-tools-write-operations

[bkatkam123]

• Is your primary need forensic/multi-server investigation or day-to-day multi-project usage?
  we have 2 master servers, I would expect it should behave as p4 CLI does, based on the workspace, I would say day-to-day usage.

• Do you need concurrent access to multiple servers/workspaces, or is switching context sufficient?
  I would just pass on workspace directory and expect it to work similar to p4 CLI

Option 2 is still static
Option 3: I don't think there is workspace switching tool

[MohitPerforce]

Thanks for the clarification. We understand your use case is day-to-day multi-project usage, with an expectation that MCP behaves similarly to the p4 CLI by inferring the active server/workspace from the workspace directory.

To clarify, MCP does support workspace switching today via the modify_workspaces tool, including a switch option (as shown in the screenshot below and documented in the README). That said, we understand your preference is for implicit, directory-based context switching rather than an explicit tool-driven switch.

We’ll take this feedback into account as we evaluate next steps.

[MohitPerforce]

@bkatkam123
The current setup already allows working with multiple Perforce servers and workspaces by defining multiple MCP server entries in the configuration, each with its own P4PORT and P4CLIENT. These can then be selected by name during usage, enabling separation between environments such as production and development.
This enables users to switch context by selecting the appropriate configured server, effectively supporting multi-server workflows.

{
"servers": {

    "p4-production": {
        "command": "/path/to/p4-mcp-server",
        "args": [],
        "env": {
            "P4PORT": "perforce:1666",
            "P4USER": "myuser",
            "P4CLIENT": "client1",
        }
    },

    "p4-development": {
        "command": "/path/to/p4-mcp-server",
        "args": [],
        "env": {
            "P4PORT": "perforce:1666",
            "P4USER": "myuser",
            "P4CLIENT": "client2",
        }
    }
    
},
"inputs": []

}

Given this, we’ll close this issue for now.