Structured text diff (line-by-line with unified output) via @mukundakatta/diff-mcp (npx) — 1 tool, Myers LCS, sub-ms

Recipe: Structured text diff via @mukundakatta/diff-mcp

Install & launch

npm install @mukundakatta/diff-mcp
# Server binary: node_modules/.bin/mcp-diff
# Or via npx: npx @mukundakatta/diff-mcp
# Transport: stdio, MCP JSON-RPC

Tool: diff

Params: a (string, original text), b (string, new text), unified (boolean, default false — include unified-diff render)

Returns: JSON with entries array of {op: "eq"|"add"|"del", text} and optionally a unified string with +/-/ prefixed lines.

Trace 1 — Code diff with unified output

→ tools/call  diff  { a: "function greet(name) {\n  console.log('Hello ' + name);\n  return true;\n}", b: "function greet(name) {\n  console.log(`Hello ${name}!`);\n  return true;\n}\nexport default greet;", unified: true }
← {
    "entries": [
      {"op":"eq","text":"function greet(name) {"},
      {"op":"del","text":"  console.log('Hello ' + name);"},
      {"op":"add","text":"  console.log(`Hello ${name}!`);"},
      {"op":"eq","text":"  return true;"},
      {"op":"eq","text":"}"},
      {"op":"add","text":"export default greet;"}
    ],
    "unified": " function greet(name) {\n-  console.log('Hello ' + name);\n+  console.log(`Hello ${name}!`);\n   return true;\n }\n+export default greet;"
  }
  latency: 1ms

Trace 2 — Prose diff

→ tools/call  diff  { a: "The quick brown fox\njumps over\nthe lazy dog", b: "The slow brown fox\nleaps over\nthe sleepy dog", unified: true }
← {
    "entries": [
      {"op":"del","text":"The quick brown fox"},
      {"op":"del","text":"jumps over"},
      {"op":"del","text":"the lazy dog"},
      {"op":"add","text":"The slow brown fox"},
      {"op":"add","text":"leaps over"},
      {"op":"add","text":"the sleepy dog"}
    ],
    "unified": "-The quick brown fox\n-jumps over\n-the lazy dog\n+The slow brown fox\n+leaps over\n+the sleepy dog"
  }
  latency: 1ms

Trace 3 — Structured-only (no unified)

→ tools/call  diff  { a: "line1\nline2\nline3", b: "line1\nline2-modified\nline3\nline4" }
← {
    "entries": [
      {"op":"eq","text":"line1"},
      {"op":"del","text":"line2"},
      {"op":"add","text":"line2-modified"},
      {"op":"eq","text":"line3"},
      {"op":"add","text":"line4"}
    ]
  }
  latency: 0ms

Notes

• Uses Myers LCS algorithm (O(n*m) DP). Fine for typical file sizes; chunk first for very large inputs.
• Line-level granularity only (no word/char mode despite initial description).
• unified: false (default) omits the rendered string — use structured entries for programmatic consumption, unified: true for human-readable output.
• Sub-millisecond latency on all calls after JIT warmup.