Skip to main content
Glama
michielinksee

Linksee Memory

by michielinksee
OverviewSchemaRelated ServersScoreDiscussions
JavaScript
Local

read_smart

Reads files with diff-only caching, returning full content first and then only changes or 'unchanged' status to save tokens on re-reads.

Instructions

Read a file with diff-only caching. Returns: (1) full content + chunk metadata on first read, (2) "unchanged" + cached chunk list (~50 tokens) if mtime matches, (3) "unchanged_content" if mtime changed but sha256 matches (touched but not modified), (4) changed chunks with content + unchanged chunks as metadata-only if the file was truly modified. Use INSTEAD of Read for files you have read before — saves 50%+ tokens on re-reads.

Input Schema

TableJSON Schema
NameRequiredDescriptionDefault
pathYesAbsolute file path
forceNoIf true, return full content regardless of cache state

Implementation Reference

  • src/mcp/read-smart.ts:38-177 (handler)
    Main handler function for the read_smart tool. Implements diff-only caching: returns full content on first read, 'unchanged' metadata on re-read if file unchanged, or only changed chunks + unchanged summary when file modified. Saves ~50%+ tokens on re-reads.
    export function handleReadSmart(
  db: Database.Database,
  args: { path: string; force?: boolean }
): string {
  const { path, force = false } = args;

  let stat;
  try {
    stat = statSync(path);
  } catch (e: any) {
    return JSON.stringify({ ok: false, error: `File not found: ${path}` });
  }

  const mtime = Math.floor(stat.mtimeMs / 1000);
  const size = stat.size;

  const prior = db.prepare('SELECT * FROM file_snapshots WHERE path = ?').get(path) as SnapshotRow | undefined;

  // --- CASE A: first read or force ---
  if (!prior || force) {
    const content = readFileSync(path, 'utf8');
    const fileHash = hashFile(content);
    const chunks = chunkFile(path, content);
    const chunkMeta = chunks.map(toMeta);

    db.prepare(
      `INSERT INTO file_snapshots (path, content_hash, mtime, size_bytes, chunks, last_read_at, read_count)
       VALUES (?, ?, ?, ?, ?, unixepoch(), 1)
       ON CONFLICT(path) DO UPDATE SET
         content_hash = excluded.content_hash,
         mtime = excluded.mtime,
         size_bytes = excluded.size_bytes,
         chunks = excluded.chunks,
         last_read_at = unixepoch(),
         read_count = read_count + 1`
    ).run(path, fileHash, mtime, size, JSON.stringify(chunkMeta));

    return JSON.stringify({
      ok: true,
      status: force ? 'forced_full' : 'first_read',
      path,
      content,
      chunks: chunkMeta,
      bytes: size,
      tokens_approx: estimateTokens(content),
      tokens_saved: 0,
    });
  }

  // --- CASE B: mtime unchanged → content guaranteed unchanged (fast path) ---
  if (prior.mtime === mtime) {
    db.prepare('UPDATE file_snapshots SET last_read_at = unixepoch(), read_count = read_count + 1 WHERE path = ?').run(path);

    const storedChunks = JSON.parse(prior.chunks) as StoredChunkMeta[];
    const factRows = db.prepare('SELECT fact, layer, chunk_hash FROM file_facts WHERE file_path = ?').all(path);

    // Token savings = what a full read would have cost
    const savedTokens = Math.round(size * TOKENS_PER_CHAR);

    return JSON.stringify({
      ok: true,
      status: 'unchanged',
      path,
      last_read_at: new Date(prior.last_read_at * 1000).toISOString(),
      chunk_count: storedChunks.length,
      chunks: storedChunks,
      file_facts: factRows,
      tokens_saved: savedTokens,
      note: 'File unchanged since last read. Call with force:true if full content is needed.',
    });
  }

  // --- CASE C: mtime changed → compute hash, maybe false alarm ---
  const content = readFileSync(path, 'utf8');
  const fileHash = hashFile(content);

  if (fileHash === prior.content_hash) {
    db.prepare('UPDATE file_snapshots SET mtime = ?, last_read_at = unixepoch(), read_count = read_count + 1 WHERE path = ?').run(mtime, path);
    return JSON.stringify({
      ok: true,
      status: 'unchanged_content',
      path,
      note: 'mtime changed but sha256 identical (file was touched but not modified).',
      tokens_saved: Math.round(size * TOKENS_PER_CHAR),
    });
  }

  // --- CASE D: real diff ---
  const newChunks = chunkFile(path, content);
  const oldChunks = JSON.parse(prior.chunks) as StoredChunkMeta[];
  const oldById = new Map(oldChunks.map((c) => [c.id, c]));

  const changedChunks: Array<{ id: string; kind: string; status: 'added' | 'modified'; start_line: number; end_line: number; content: string }> = [];
  const unchangedChunks: StoredChunkMeta[] = [];
  const seenIds = new Set<string>();

  for (const c of newChunks) {
    seenIds.add(c.id);
    const prev = oldById.get(c.id);
    if (!prev) {
      changedChunks.push({ id: c.id, kind: c.kind, status: 'added', start_line: c.start_line, end_line: c.end_line, content: c.content });
    } else if (prev.hash !== c.hash) {
      changedChunks.push({ id: c.id, kind: c.kind, status: 'modified', start_line: c.start_line, end_line: c.end_line, content: c.content });
    } else {
      unchangedChunks.push({ id: c.id, kind: c.kind, start_line: c.start_line, end_line: c.end_line, hash: c.hash });
    }
  }

  const removedChunks = oldChunks
    .filter((c) => !seenIds.has(c.id))
    .map((c) => ({ id: c.id, kind: c.kind, prev_lines: `${c.start_line}-${c.end_line}` }));

  const newChunkMeta = newChunks.map(toMeta);
  db.prepare(
    `UPDATE file_snapshots SET content_hash = ?, mtime = ?, size_bytes = ?, chunks = ?, last_read_at = unixepoch(), read_count = read_count + 1 WHERE path = ?`
  ).run(fileHash, mtime, size, JSON.stringify(newChunkMeta), path);

  const fullTokens = estimateTokens(content);
  const returnedTokens = changedChunks.reduce((s, c) => s + estimateTokens(c.content), 0) + 80; // ~80 for the envelope
  const savedTokens = Math.max(0, fullTokens - returnedTokens);
  const pctSaved = fullTokens > 0 ? Math.round((savedTokens / fullTokens) * 100) : 0;

  return JSON.stringify({
    ok: true,
    status: 'modified',
    path,
    changed_chunks: changedChunks,
    unchanged_chunks: unchangedChunks,
    removed_chunks: removedChunks,
    summary: {
      changed: changedChunks.length,
      unchanged: unchangedChunks.length,
      removed: removedChunks.length,
      tokens_full: fullTokens,
      tokens_returned: returnedTokens,
      tokens_saved: savedTokens,
      pct_saved: pctSaved,
    },
  });
}
  • src/mcp/server.ts:167-177 (schema)
    Input schema and tool registration definition for read_smart. Defines path (required string) and force (optional boolean) inputs with descriptions.
    {
  name: 'read_smart',
  description:
    'Read a file with diff-only caching. Returns: (1) full content + chunk metadata on first read, (2) "unchanged" + cached chunk list (~50 tokens) if mtime matches, (3) "unchanged_content" if mtime changed but sha256 matches (touched but not modified), (4) changed chunks with content + unchanged chunks as metadata-only if the file was truly modified. Use INSTEAD of Read for files you have read before — saves 50%+ tokens on re-reads.',
  inputSchema: {
    type: 'object',
    properties: {
      path: { type: 'string', description: 'Absolute file path' },
      force: { type: 'boolean', description: 'If true, return full content regardless of cache state', default: false },
    },
    required: ['path'],
  • src/mcp/server.ts:811-811 (registration)
    Tool dispatch – routes the 'read_smart' tool name from the MCP CallToolRequest to the handleReadSmart wrapper function.
    case 'read_smart': text = handleReadSmart(args); break;
  • src/lib/file-chunker.ts:44-273 (helper)
    File chunker used by read_smart to split files into semantic chunks (AST-based for TS/JS, indent-based for Python, heading-based for Markdown, fixed 100-line windows for others). Enables diff-only re-reads by computing chunk hashes.
    export function chunkFile(path: string, content: string): Chunk[] {
  const ext = extname(path).toLowerCase();
  try {
    if (['.ts', '.tsx', '.js', '.jsx', '.mjs', '.cjs'].includes(ext)) {
      return chunkTsJs(content);
    }
    if (ext === '.py') return chunkPython(content);
    if (ext === '.md' || ext === '.markdown') return chunkMarkdown(content);
  } catch {
    // parse failures fall back to fixed
  }
  return chunkFixed(content, 100);
}

// ============================================================
// TypeScript / JavaScript — AST-based
// ============================================================
function chunkTsJs(content: string): Chunk[] {
  const ast = babelParse(content, {
    sourceType: 'unambiguous',
    plugins: ['typescript', 'jsx', 'decorators-legacy'],
    errorRecovery: true,
    allowImportExportEverywhere: true,
    allowReturnOutsideFunction: true,
  });

  const chunks: Chunk[] = [];
  const importNodes: { start: number; end: number }[] = [];

  for (const node of (ast as any).program.body) {
    if (!node.loc) continue;
    const start = node.loc.start.line;
    const end = node.loc.end.line;

    if (node.type === 'ImportDeclaration') {
      importNodes.push({ start, end });
      continue;
    }

    const name = extractDeclName(node);
    const kind = mapNodeKind(node);
    const text = extractLines(content, start, end);
    chunks.push({
      id: `${kind}:${name}`,
      kind,
      start_line: start,
      end_line: end,
      hash: shortHash(text),
      content: text,
    });
  }

  if (importNodes.length > 0) {
    const first = importNodes[0].start;
    const last = importNodes[importNodes.length - 1].end;
    const text = extractLines(content, first, last);
    chunks.unshift({
      id: 'import:_block',
      kind: 'import',
      start_line: first,
      end_line: last,
      hash: shortHash(text),
      content: text,
    });
  }

  if (chunks.length === 0) return chunkFixed(content, 100);
  return chunks;
}

function extractDeclName(node: any): string {
  if (node.type === 'FunctionDeclaration') return node.id?.name ?? 'anonymous';
  if (node.type === 'ClassDeclaration') return node.id?.name ?? 'anonymous';
  if (node.type === 'VariableDeclaration') {
    const d = node.declarations?.[0];
    return d?.id?.name ?? 'anonymous';
  }
  if (node.type === 'ExportNamedDeclaration') {
    if (node.declaration) return extractDeclName(node.declaration);
    const first = node.specifiers?.[0];
    return first?.exported?.name ?? 'named_export';
  }
  if (node.type === 'ExportDefaultDeclaration') {
    if (node.declaration?.id?.name) return node.declaration.id.name;
    return 'default';
  }
  if (node.type === 'TSInterfaceDeclaration') return node.id?.name ?? 'interface';
  if (node.type === 'TSTypeAliasDeclaration') return node.id?.name ?? 'type';
  if (node.type === 'TSEnumDeclaration') return node.id?.name ?? 'enum';
  if (node.type === 'TSModuleDeclaration') return node.id?.name ?? 'module';
  return node.type;
}

function mapNodeKind(node: any): ChunkKind {
  const t = node.type;
  if (t === 'FunctionDeclaration') return 'function';
  if (t === 'ClassDeclaration') return 'class';
  if (['VariableDeclaration', 'TSInterfaceDeclaration', 'TSTypeAliasDeclaration', 'TSEnumDeclaration', 'TSModuleDeclaration'].includes(t)) return 'variable';
  if (t === 'ExportNamedDeclaration' || t === 'ExportDefaultDeclaration') {
    if (node.declaration) return mapNodeKind(node.declaration);
    return 'variable';
  }
  return 'variable';
}

// ============================================================
// Python — indent-based
// ============================================================
function chunkPython(content: string): Chunk[] {
  const lines = content.split('\n');
  const chunks: Chunk[] = [];
  let current: { start: number; name: string; kind: 'python_def' | 'python_class' } | null = null;

  const pushCurrent = (end: number) => {
    if (!current) return;
    const text = extractLines(content, current.start, end);
    chunks.push({
      id: `${current.kind}:${current.name}`,
      kind: current.kind,
      start_line: current.start,
      end_line: end,
      hash: shortHash(text),
      content: text,
    });
    current = null;
  };

  for (let i = 0; i < lines.length; i++) {
    const line = lines[i];
    if (line.startsWith(' ') || line.startsWith('\t')) continue;
    const m = line.match(/^(async\s+def|def|class)\s+(\w+)/);
    if (m) {
      pushCurrent(i); // previous ends at the line before current
      const kind: 'python_def' | 'python_class' = m[1] === 'class' ? 'python_class' : 'python_def';
      current = { start: i + 1, name: m[2], kind };
    }
  }
  pushCurrent(lines.length);

  if (chunks.length === 0) return chunkFixed(content, 100);

  if (chunks[0].start_line > 1) {
    const preambleText = extractLines(content, 1, chunks[0].start_line - 1);
    if (preambleText.trim()) {
      chunks.unshift({
        id: 'python_preamble',
        kind: 'python_preamble',
        start_line: 1,
        end_line: chunks[0].start_line - 1,
        hash: shortHash(preambleText),
        content: preambleText,
      });
    }
  }

  return chunks;
}

// ============================================================
// Markdown — h2/h3 boundaries
// ============================================================
function chunkMarkdown(content: string): Chunk[] {
  const lines = content.split('\n');
  const chunks: Chunk[] = [];
  let current: { start: number; heading: string } | null = null;

  const pushCurrent = (end: number) => {
    if (!current) return;
    const text = extractLines(content, current.start, end);
    chunks.push({
      id: `heading:${current.heading}`,
      kind: 'heading',
      start_line: current.start,
      end_line: end,
      hash: shortHash(text),
      content: text,
    });
    current = null;
  };

  for (let i = 0; i < lines.length; i++) {
    const m = lines[i].match(/^(#{2,3})\s+(.+?)\s*$/);
    if (m) {
      pushCurrent(i);
      current = { start: i + 1, heading: m[2] };
    }
  }
  pushCurrent(lines.length);

  if (chunks.length === 0) return chunkFixed(content, 100);

  if (chunks[0].start_line > 1) {
    const preamble = extractLines(content, 1, chunks[0].start_line - 1);
    if (preamble.trim()) {
      chunks.unshift({
        id: 'heading:_preamble',
        kind: 'heading',
        start_line: 1,
        end_line: chunks[0].start_line - 1,
        hash: shortHash(preamble),
        content: preamble,
      });
    }
  }

  return chunks;
}

// ============================================================
// Fixed — 100-line windows (fallback)
// ============================================================
function chunkFixed(content: string, size: number): Chunk[] {
  const lines = content.split('\n');
  const chunks: Chunk[] = [];
  if (lines.length === 0) return chunks;
  for (let i = 0; i < lines.length; i += size) {
    const slice = lines.slice(i, i + size);
    const text = slice.join('\n');
    const end = Math.min(i + size, lines.length);
    chunks.push({
      id: `lines:${i + 1}_${end}`,
      kind: 'fixed',
      start_line: i + 1,
      end_line: end,
      hash: shortHash(text),
      content: text,
    });
  }
  return chunks;
}
  • src/mcp/server.ts:789-791 (registration)
    Thin wrapper that calls the imported handleReadSmart implementation from read-smart.ts, passing the database instance and parsed arguments.
    function handleReadSmart(args: any): string {
  return handleReadSmartImpl(db, { path: args.path, force: args.force });
}

Tool Definition Quality

A4.6/5.0
Behavior5/5

Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?

With no annotations provided, the description fully discloses the caching behavior and all four possible return states based on file modification status, offering complete transparency.

Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.

Conciseness4/5

Is the description appropriately sized, front-loaded, and free of redundancy?

The description is moderately concise and well-structured with numbered return cases. While informative, it could be slightly more compact without losing clarity.

Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.

Completeness5/5

Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?

Despite lacking an output schema, the description thoroughly explains all four possible return types and the effect of the 'force' parameter, making it complete for an agent to understand the tool's behavior.

Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.

Parameters3/5

Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?

Input schema has 100% coverage with descriptions for both parameters. The tool description does not add additional semantic value beyond what the schema already provides, so a baseline score of 3 is appropriate.

Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.

Purpose5/5

Does the description clearly state what the tool does and how it differs from similar tools?

The description clearly states the tool reads a file with diff-only caching and lists four distinct return scenarios. It explicitly distinguishes itself from an alternative 'Read' tool by advising when to use it instead.

Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.

Usage Guidelines5/5

Does the description explain when to use this tool, when not to, or what alternatives exist?

The description explicitly states 'Use INSTEAD of Read for files you have read before — saves 50%+ tokens on re-reads', providing clear guidance on when to use this tool vs. an alternative.

Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.

Install Server

Other Tools

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/michielinksee/linksee-memory'

If you have feedback or need assistance with the MCP directory API, please join our Discord server