MCP Server Error: Fails to parse issue content (ZodError)

[davidorban]

I'm encountering an error when using the linear-mcp-server via Roo Code.

Environment:

• OS: Windows 10
• Client: Roo Code (VS Code Extension)
• Server Configuration: I added the server configuration manually to my project's .roo/mcp.json file using the JSON generated by Smithery. The server is correctly recognized and listed by Roo Code. My configuration looks like this:

  "linear-mcp-server": {
    "command": "cmd", // Using cmd /c for Windows consistency
    "args": [
      "/c",
      "npx",
      "-y",
      "@smithery/cli@latest",
      "run",
      "linear-mcp-server",
      "--key",
      "YOUR_LINEAR_API_KEY_HERE" // Masked for security
    ]
  }

• My Linear API key is confirmed to be valid.

Steps to Reproduce:

1. Attempted to use the linear_search_issues tool via Roo Code (e.g., searching for issues with status "Todo").
2. Attempted to use the linear_get_user_issues tool via Roo Code.

Expected Result:

I expected the tools to return a list of my Linear issues based on the criteria.

Actual Result:

Both tool calls failed consistently, returning a server-side error originating from the MCP server. The error indicates a failure in parsing the response data, specifically related to the issue content.

The core error message received from the server is a ZodError:

ZodError: Invalid input

Path: ["content", 0]
Union Errors involving:
Path: ["content", 0, "text"], Message: "Expected string, received object"
Path: ["content", 0, "type"], Message: "Invalid literal value, expected 'image'/'resource'" (or similar details indicating a mismatch with expected simple text)

(Full stack traces were provided in the Roo Code interface, but the key issue seems to be the schema expecting content[0].text to be a string but receiving an object instead).

Hypothesis:

It seems the server's schema validation (using Zod) expects the content.text field within an issue's data structure to always be a simple string. However, it appears that when an issue description or comment contains rich text, links, images, or other non-plain-text elements, the Linear API returns a more complex object structure for that content. The current server schema doesn't seem equipped to handle this object structure, leading to the parsing failure.

Could you please investigate if the schema definition for issue content needs to be updated to handle potentially complex object structures instead of just plain strings?

Thanks!