search_text

[semantic]

Searches for a literal string or regex pattern (non-semantic) over the documents the workspace has loaded - source files plus any additional files and analyzer config a project declares - as the compiler last ingested them - the file watcher keeps that in step with disk, and sync/reload force it, so results track saved files rather than an unsaved editor buffer. Files no loaded project references are out of scope: for those, add scout as its own MCP server and use its find tool. Requires a completed workspace load.

Why this tool exists

Plain text still matters for comments, config-like code, string literals, and broad repo reconnaissance.

How it helps the agent

It gives the model an intentional text-search path inside the same loaded-root workflow, instead of mixing semantic and shell evidence casually.

Try it locally in the Glider playground

Parameters

Name	Type	Required	Description
query	string	No	Search query (literal string by default). Preferred argument name.
pattern	string	No	Backward-compatible alias for query.
useRegex	boolean	No	Treat pattern as a regex (.NET dialect). Default is false.
caseSensitive	boolean	No	Case-sensitive match. Default is false.
scope	json	No	Optional search scope.
skip	number	No	Pagination offset. Default is 0.
take	number	No	Pagination size. Default is 200.
maxLineTextChars	number	No	Max characters for lineText. Use 0 for unlimited. Default is 200.
pathStyle	string	No	Path style: 'absolute' (default) or 'relative' (to solution root).

Examples

Find TODO comments

Request 
{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "search_text",
    "arguments": {
      "query": "TODO"
    }
  }
}

Response 
{
  "success": true,
  "data": {
    "coverage": "loadedDocuments",
    "pattern": "TODO",
    "useRegex": false,
    "caseSensitive": false,
    "scopeUsed": {
      "mode": "solution"
    },
    "matchCount": 2,
    "projectsSearched": 2,
    "totalProjectsInWorkspace": 2,
    "documentsMatchedScope": 120,
    "documentsSearched": 120,
    "documentsWithMatches": 1,
    "documentsUnreadable": 0,
    "totalDocumentsInWorkspace": 120,
    "paging": {
      "skip": 0,
      "take": 200,
      "returned": 2,
      "total": 2
    },
    "matches": [
      {
        "filePath": "/path/to/Program.cs",
        "lineNumber": 12,
        "column": 1,
        "lineText": "// TODO: refactor this",
        "projectName": "MyProject"
      }
    ],
    "hints": {
      "nextSteps": [
        "Use get_symbol_at_position if the match is an identifier and you want a symbolKey",
        "Use search_symbols for semantic discovery once you know a symbol name"
      ],
      "scopeNote": null,
      "noMatchNote": null,
      "unreadableNote": null
    }
  },
  "meta": {
    "durationMs": 123,
    "cancelled": false,
    "timedOut": false,
    "timeoutMs": 1200000
  },
  "error": null
}

Response Notes

Returns matching text occurrences with paging. data.coverage always reports 'loadedDocuments' - a standing statement of scope, not a backend discriminator. Before reading a zero-hit result as proof the code is absent, check how much was actually searched: compare documentsSearched with documentsMatchedScope, and check documentsUnreadable. A scope that matched no documents is reported in hints.scopeNote, and an unreadable-document count in hints.unreadableNote. A zero-hit search fills hints.noMatchNote, which names how many documents were searched and which file kinds that scope excludes. Text outside the loaded workspace is not covered at all - use scout's find for that.