Tabnas MCP server
{
"mcpServers": {
"tabnas": {
"command": "npx",
"args": [
"-y",
"@tabnas/mcp@latest"
]
}
}
}{
"mcpServers": {
"tabnas": {
"command": "npx",
"args": [
"-y",
"@tabnas/mcp@latest"
]
}
}
}{
"mcpServers": {
"tabnas": {
"command": "npx",
"args": [
"-y",
"@tabnas/mcp@latest"
]
}
}
}{
"servers": {
"tabnas": {
"command": "npx",
"args": [
"-y",
"@tabnas/mcp@latest"
]
}
}
}{
"mcpServers": {
"tabnas": {
"command": "npx",
"args": [
"-y",
"@tabnas/mcp@latest"
]
}
}
}file .mcp.json (in your project root)file claude_desktop_config.jsonfile ~/.cursor/mcp.jsonfile .vscode/mcp.jsonfile ~/.codeium/windsurf/mcp_config.jsoninstall npx -y @tabnas/mcpversion 0.1.15
What it does
Tabnas exposes a parsing engine through seven tools for grammar validation, parsing, testing, and comparison. It serves developers who work with custom grammars and need to parse, validate, and test input against grammar specifications.
- Parse input text against a custom grammar and get back a parse tree or a diagnostic error.
- Test a grammar definition against a suite of TSV fixtures to verify expected behavior.
- Compare two grammar versions to detect whether the new grammar accepts the same inputs and produces identical trees as the old one.
Tools this server exposes
| Tool | Does |
|---|---|
| parse | Parse input text against a grammar and return the parse tree or a diagnostic error. |
| validate_grammar | Validate a serialized grammar specification for structural correctness and security. |
| explain_parse_error | Explain why a given input failed to parse, with reference to the error code registry. |
| test_grammar | Run a grammar against TSV fixtures and report pass/fail results for each test row. |
| list_plugins | List all available grammar plugins. |
| describe_plugin | Return the full descriptor object for a specific grammar plugin. |
| compare_grammars | Compare two grammar definitions to check backward compatibility and tree equivalence. |
Before you enable it
An MCP server runs with whatever access you give it. Start with a development account rather than production, and check which tools the server exposes before you let an agent call them unattended.