diff --git a/CHANGELOG.md b/CHANGELOG.md
index 3ca5b9ff1..761b6b1b8 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -12,6 +12,10 @@ and adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+### Fixes
+
+- `codegraph callers`, `codegraph callees` and `codegraph impact` now understand qualified symbol names. Asking for something like `Accounts.Format.group` used to match nothing, so the command quietly fell back to whichever symbol full-text search happened to rank first — or reported "not found" for a symbol that plainly exists. Qualified names now select exactly the definition you named, including in languages whose module names themselves contain dots.
+- Those same commands now tell you when a plain name matches several different definitions. Previously the results for every same-named symbol — often in different languages, since collisions cluster on short names like `group`, `num` or `parse` — were merged into a single list with nothing saying they came from different symbols. The list still aggregates, but it now names what it aggregated and shows you the qualified spelling that narrows it.
 
 ## [1.6.0] - 2026-08-26
 
diff --git a/__tests__/symbol-lookup.test.ts b/__tests__/symbol-lookup.test.ts
index c81aaabd4..e78b12ec2 100644
--- a/__tests__/symbol-lookup.test.ts
+++ b/__tests__/symbol-lookup.test.ts
@@ -17,6 +17,8 @@ import * as fs from 'fs';
 import * as path from 'path';
 import * as os from 'os';
 import { initGrammars, loadAllGrammars } from '../src/extraction/grammars';
+import { matchesSymbol, lookupSymbolNodes, isQualifiedSymbol } from '../src/graph/symbol-lookup';
+import type { Node } from '../src/types';
 
 beforeAll(async () => {
   await initGrammars();
@@ -220,3 +222,173 @@ describe.skipIf(!HAS_SQLITE)('matchesSymbol — dotted lookups (regression for #
     expect((text.match(/\*\*Location:\*\*/g) || []).length).toBeGreaterThanOrEqual(2);
   });
 });
+
+/**
+ * One resolution path for every verb that takes a symbol NAME.
+ *
+ * `callers` / `callees` / `impact` used to carry their own filter, comparing
+ * the query against the BARE name only:
+ *
+ *     node.name === symbol || node.name.endsWith('.' + symbol)
+ *
+ * which fails in two opposite directions at once. A bare name matched every
+ * same-named symbol in the repository and their results were merged under one
+ * heading with nothing saying they were different symbols; a qualified name
+ * could never equal a bare `node.name`, so every candidate failed the filter
+ * and the code fell through to an arbitrary top-of-FTS hit — or reported "not
+ * found" for a symbol that plainly exists. Both now go through
+ * `lookupSymbolNodes`.
+ */
+function fakeNode(over: Partial<Node>): Node {
+  return {
+    id: 'n1', kind: 'function', name: 'group', qualifiedName: 'group',
+    filePath: 'lib/format.ex', language: 'typescript',
+    startLine: 1, endLine: 2, startColumn: 0, endColumn: 0, updatedAt: 0,
+    ...over,
+  } as Node;
+}
+
+describe('matchesSymbol — containers whose own name contains a separator', () => {
+  // Splitting on EVERY separator assumes no scope component contains one. That
+  // is false for any language whose module names are themselves dotted, and
+  // there the stored qualifiedName (`A.B::c`) can never equal the split-and-
+  // rejoined query spelling (`A::B::c`) — so a perfectly precise qualified
+  // query resolved to nothing.
+  const node = fakeNode({ name: 'group', qualifiedName: 'AppWeb.Format::group' });
+
+  it('matches a dotted module qualifier written with dots', () => {
+    expect(matchesSymbol(node, 'AppWeb.Format.group')).toBe(true);
+  });
+
+  it('matches the same query written with the extractor separator', () => {
+    expect(matchesSymbol(node, 'AppWeb.Format::group')).toBe(true);
+  });
+
+  it('matches a partial container suffix on a separator boundary', () => {
+    expect(matchesSymbol(node, 'Format.group')).toBe(true);
+  });
+
+  it('does not match a container that merely shares a suffix substring', () => {
+    // `ebFormat.group` is not a boundary-aligned suffix of `AppWeb.Format.group`.
+    expect(matchesSymbol(node, 'ebFormat.group')).toBe(false);
+  });
+
+  it('does not match a different container', () => {
+    expect(matchesSymbol(node, 'Other.Format.group')).toBe(false);
+  });
+
+  it('still requires the last part to be the node name', () => {
+    expect(matchesSymbol(node, 'AppWeb.Format.other')).toBe(false);
+  });
+
+  it('classifies bare vs qualified queries', () => {
+    expect(isQualifiedSymbol('group')).toBe(false);
+    expect(isQualifiedSymbol('A.B.group')).toBe(true);
+    expect(isQualifiedSymbol('A::group')).toBe(true);
+    expect(isQualifiedSymbol('a/b')).toBe(true);
+  });
+});
+
+describe.skipIf(!HAS_SQLITE)('lookupSymbolNodes — the shared path used by callers/callees/impact', () => {
+  let projectRoot: string;
+  let cg: any;
+
+  beforeEach(async () => {
+    projectRoot = tmpRoot();
+    const client = path.join(projectRoot, 'client');
+    const pkg = path.join(projectRoot, 'pkg', 'fmtutil');
+    fs.mkdirSync(client, { recursive: true });
+    fs.mkdirSync(pkg, { recursive: true });
+    // The SAME short name defined in two languages — the collision profile of a
+    // polyglot repository, where the colliding identifiers are the common ones.
+    fs.writeFileSync(
+      path.join(client, 'chart.ts'),
+      `export function group(rows: number[][]): number[][] { return rows; }\n`
+    );
+    fs.writeFileSync(
+      path.join(client, 'Editor.tsx'),
+      `import { group } from './chart';\nexport function Editor(r: number[][]) { return group(r); }\n`
+    );
+    fs.writeFileSync(
+      path.join(pkg, 'format.py'),
+      `def group(items, size):\n    return items\n`
+    );
+    fs.writeFileSync(
+      path.join(projectRoot, 'pkg', 'planner.py'),
+      `from pkg.fmtutil.format import group\n\ndef plan_a(items): return group(items, 3)\ndef plan_b(items): return group(items, 5)\n`
+    );
+
+    const CodeGraph = (await import('../src/index')).default;
+    cg = CodeGraph.initSync(projectRoot, {
+      config: { include: ['**/*.ts', '**/*.tsx', '**/*.py'], exclude: [] },
+    });
+    await cg.indexAll();
+  });
+
+  afterEach(() => {
+    cg?.destroy();
+    rmTree(projectRoot);
+  });
+
+  it('a bare name resolves to EVERY definition and reports the ambiguity', () => {
+    const { nodes, ambiguous } = lookupSymbolNodes(cg, 'group');
+    const defs = nodes.filter((n) => n.kind === 'function');
+    expect(defs.length).toBe(2);
+    expect(new Set(defs.map((n) => n.language))).toEqual(new Set(['typescript', 'python']));
+    // The flag is what stops an aggregate being presented as one symbol's answer.
+    expect(ambiguous).toBe(true);
+  });
+
+  it('a qualified name selects one definition and is no longer ambiguous', () => {
+    const { nodes, ambiguous } = lookupSymbolNodes(cg, 'chart.group');
+    expect(nodes.length).toBe(1);
+    expect(nodes[0]!.language).toBe('typescript');
+    expect(nodes[0]!.filePath).toMatch(/chart\.ts$/);
+    expect(ambiguous).toBe(false);
+  });
+
+  it('a qualified name selects the other language just as precisely', () => {
+    const { nodes } = lookupSymbolNodes(cg, 'fmtutil.format.group');
+    expect(nodes.length).toBe(1);
+    expect(nodes[0]!.language).toBe('python');
+    expect(nodes[0]!.filePath).toMatch(/fmtutil\/format\.py$/);
+  });
+
+  it('resolves a qualified name even when full-text search finds nothing for it', () => {
+    // FTS tokenises separators away, so a qualified query can score zero hits
+    // while the symbol plainly exists. Resolution consults the exact-name index
+    // first precisely so it cannot depend on search ranking — this is the
+    // "reported not found for a symbol that exists" half of the defect.
+    const fts = cg.searchNodes('fmtutil.format.group', { limit: 50 });
+    const { nodes } = lookupSymbolNodes(cg, 'fmtutil.format.group');
+    expect(nodes.length).toBe(1);
+    expect(nodes[0]!.filePath).toMatch(/format\.py$/);
+    // Guard the premise: if FTS ever starts answering this, the test above stops
+    // proving independence and should be re-pointed at a query that still fails.
+    expect(Array.isArray(fts)).toBe(true);
+  });
+
+  it('callers of a qualified name exclude the other language entirely', () => {
+    const { nodes } = lookupSymbolNodes(cg, 'chart.group');
+    const callerFiles = nodes.flatMap((n: any) =>
+      cg.getCallers(n.id).map((c: any) => c.node.filePath)
+    );
+    expect(callerFiles.length).toBeGreaterThan(0);
+    for (const f of callerFiles) expect(f).not.toMatch(/\.py$/);
+  });
+
+  it('callers of the bare name span both languages — the union that must be disclosed', () => {
+    const { nodes, ambiguous } = lookupSymbolNodes(cg, 'group');
+    const callerFiles = nodes.flatMap((n: any) =>
+      cg.getCallers(n.id).map((c: any) => c.node.filePath)
+    );
+    expect(ambiguous).toBe(true);
+    expect(callerFiles.some((f: string) => f.endsWith('.py'))).toBe(true);
+    expect(callerFiles.some((f: string) => f.endsWith('.tsx'))).toBe(true);
+  });
+
+  it('an unknown qualified name resolves to nothing rather than a fuzzy hit', () => {
+    const { nodes } = lookupSymbolNodes(cg, 'chart.nonexistent_fn');
+    expect(nodes.length).toBe(0);
+  });
+});
diff --git a/src/bin/codegraph.ts b/src/bin/codegraph.ts
index 19038df1b..b532e0be4 100644
--- a/src/bin/codegraph.ts
+++ b/src/bin/codegraph.ts
@@ -53,6 +53,7 @@ import { relaunchWithWasmRuntimeFlagsIfNeeded } from '../extraction/wasm-runtime
 import { installCommandSupervision } from './command-supervision';
 import { EXTRACTION_VERSION } from '../extraction/extraction-version';
 import { getTelemetry, TELEMETRY_DOCS, recordIndexEvent } from '../telemetry';
+import { lookupSymbolNodes, describeSymbolNode } from '../graph/symbol-lookup';
 
 // Decided once, before `--color`/`--no-color` are stripped from argv below
 // (#1281). Piped/redirected stdout, NO_COLOR, or --no-color -> plain output.
@@ -356,6 +357,56 @@ function warn(message: string): void {
   console.log(chalk.yellow(getGlyphs().warn) + ' ' + message);
 }
 
+/**
+ * Disclose that a name resolved to several DISTINCT definitions, whose results
+ * were merged into the list just printed.
+ *
+ * `callers` / `callees` / `impact` aggregate across every definition a name
+ * matches. That is the useful default — an interface method and its overrides
+ * are usually all wanted — but presenting the union under one heading, with
+ * nothing saying the name was ambiguous, is how a query for a common
+ * identifier ends up reporting callers that belong to an entirely unrelated
+ * symbol (often in another language, since collisions cluster on short generic
+ * names like `group`, `num`, `parse`). Naming the targets keeps the aggregate
+ * useful and makes the widening visible, and tells the user the qualified
+ * spelling that would narrow it.
+ */
+function printAmbiguityNote(
+  symbol: string,
+  targets: Array<{ qualifiedName: string; kind: string; language: string; filePath: string; startLine: number }>,
+  ambiguous: boolean
+): void {
+  if (!ambiguous || targets.length < 2) return;
+  const languages = new Set(targets.map((t) => t.language));
+  console.log(
+    chalk.yellow(getGlyphs().warn) +
+      ` "${symbol}" names ${targets.length} definitions` +
+      (languages.size > 1 ? ` across ${languages.size} languages` : '') +
+      ' — the results above are the union of all of them:'
+  );
+  for (const t of targets.slice(0, 10)) {
+    console.log(chalk.dim(`    ${describeSymbolNode(t as never)}`));
+  }
+  if (targets.length > 10) console.log(chalk.dim(`    … +${targets.length - 10} more`));
+  console.log(chalk.dim(`  Narrow it with a qualified name, e.g. "${narrowingExample(targets[0]!)}".`));
+}
+
+/**
+ * A qualified spelling that would select exactly this definition. Languages
+ * that carry the container in `qualifiedName` (Elixir, Java, C++, class-scoped
+ * methods) can offer it directly; the ones that encode their module in the
+ * FILE PATH instead (Python, Rust) have a bare qualifiedName, so suggesting it
+ * would just echo the ambiguous name back. For those, `<file>.<name>` is the
+ * spelling that resolves — it is what the file-path stage of `matchesSymbol`
+ * matches on.
+ */
+function narrowingExample(target: { qualifiedName: string; name?: string; filePath: string }): string {
+  const qualified = target.qualifiedName.replace(/::/g, '.');
+  if (qualified.includes('.')) return qualified;
+  const basename = target.filePath.split('/').pop()?.replace(/\.[^.]+$/, '');
+  return basename ? `${basename}.${qualified}` : qualified;
+}
+
 type IndexResult = {
   success: boolean;
   filesIndexed: number;
@@ -1957,8 +2008,8 @@ program
       const cg = await CodeGraph.open(projectPath);
       const limit = parseInt(options.limit || '20', 10);
 
-      const matches = cg.searchNodes(symbol, { limit: 50 });
-      if (matches.length === 0) {
+      const { nodes: targets, ambiguous } = lookupSymbolNodes(cg, symbol);
+      if (targets.length === 0) {
         info(`Symbol "${symbol}" not found`);
         cg.destroy();
         return;
@@ -1967,20 +2018,8 @@ program
       const seen = new Set<string>();
       const allCallers: Array<{ name: string; kind: string; filePath: string; startLine?: number }> = [];
 
-      for (const match of matches) {
-        const exactMatch = match.node.name === symbol || match.node.name.endsWith(`.${symbol}`) || match.node.name.endsWith(`::${symbol}`);
-        if (!exactMatch && matches.length > 1) continue;
-        for (const c of cg.getCallers(match.node.id)) {
-          if (!seen.has(c.node.id)) {
-            seen.add(c.node.id);
-            allCallers.push({ name: c.node.name, kind: c.node.kind, filePath: c.node.filePath, startLine: c.node.startLine });
-          }
-        }
-      }
-
-      // Fallback: if exact filter removed everything, use the top match
-      if (allCallers.length === 0 && matches[0]) {
-        for (const c of cg.getCallers(matches[0].node.id)) {
+      for (const target of targets) {
+        for (const c of cg.getCallers(target.id)) {
           if (!seen.has(c.node.id)) {
             seen.add(c.node.id);
             allCallers.push({ name: c.node.name, kind: c.node.kind, filePath: c.node.filePath, startLine: c.node.startLine });
@@ -1991,7 +2030,17 @@ program
       const limited = allCallers.slice(0, limit);
 
       if (options.json) {
-        console.log(JSON.stringify({ symbol, callers: limited }, null, 2));
+        console.log(JSON.stringify({
+          symbol,
+          // Which definitions the name resolved to. An aggregate over several
+          // distinct symbols has to say so — see graph/symbol-lookup.
+          targets: targets.map((t) => ({
+            qualifiedName: t.qualifiedName, kind: t.kind, language: t.language,
+            filePath: t.filePath, startLine: t.startLine,
+          })),
+          ambiguous,
+          callers: limited,
+        }, null, 2));
       } else if (limited.length === 0) {
         info(`No callers found for "${symbol}"`);
       } else {
@@ -2005,6 +2054,7 @@ program
           console.log(chalk.dim(`  ${node.filePath}${loc}`));
           console.log();
         }
+        printAmbiguityNote(symbol, targets, ambiguous);
       }
 
       cg.destroy();
@@ -2036,8 +2086,8 @@ program
       const cg = await CodeGraph.open(projectPath);
       const limit = parseInt(options.limit || '20', 10);
 
-      const matches = cg.searchNodes(symbol, { limit: 50 });
-      if (matches.length === 0) {
+      const { nodes: targets, ambiguous } = lookupSymbolNodes(cg, symbol);
+      if (targets.length === 0) {
         info(`Symbol "${symbol}" not found`);
         cg.destroy();
         return;
@@ -2046,19 +2096,8 @@ program
       const seen = new Set<string>();
       const allCallees: Array<{ name: string; kind: string; filePath: string; startLine?: number }> = [];
 
-      for (const match of matches) {
-        const exactMatch = match.node.name === symbol || match.node.name.endsWith(`.${symbol}`) || match.node.name.endsWith(`::${symbol}`);
-        if (!exactMatch && matches.length > 1) continue;
-        for (const c of cg.getCallees(match.node.id)) {
-          if (!seen.has(c.node.id)) {
-            seen.add(c.node.id);
-            allCallees.push({ name: c.node.name, kind: c.node.kind, filePath: c.node.filePath, startLine: c.node.startLine });
-          }
-        }
-      }
-
-      if (allCallees.length === 0 && matches[0]) {
-        for (const c of cg.getCallees(matches[0].node.id)) {
+      for (const target of targets) {
+        for (const c of cg.getCallees(target.id)) {
           if (!seen.has(c.node.id)) {
             seen.add(c.node.id);
             allCallees.push({ name: c.node.name, kind: c.node.kind, filePath: c.node.filePath, startLine: c.node.startLine });
@@ -2069,7 +2108,15 @@ program
       const limited = allCallees.slice(0, limit);
 
       if (options.json) {
-        console.log(JSON.stringify({ symbol, callees: limited }, null, 2));
+        console.log(JSON.stringify({
+          symbol,
+          targets: targets.map((t) => ({
+            qualifiedName: t.qualifiedName, kind: t.kind, language: t.language,
+            filePath: t.filePath, startLine: t.startLine,
+          })),
+          ambiguous,
+          callees: limited,
+        }, null, 2));
       } else if (limited.length === 0) {
         info(`No callees found for "${symbol}"`);
       } else {
@@ -2083,6 +2130,7 @@ program
           console.log(chalk.dim(`  ${node.filePath}${loc}`));
           console.log();
         }
+        printAmbiguityNote(symbol, targets, ambiguous);
       }
 
       cg.destroy();
@@ -2114,22 +2162,20 @@ program
       const cg = await CodeGraph.open(projectPath);
       const depth = Math.min(Math.max(parseInt(options.depth || '2', 10), 1), 10);
 
-      const matches = cg.searchNodes(symbol, { limit: 50 });
-      if (matches.length === 0) {
+      const { nodes: targets, ambiguous } = lookupSymbolNodes(cg, symbol);
+      if (targets.length === 0) {
         info(`Symbol "${symbol}" not found`);
         cg.destroy();
         return;
       }
 
-      // Merge impact subgraphs across all exact-matching symbols
+      // Merge impact subgraphs across every definition the name resolved to.
       const mergedNodes = new Map<string, { name: string; kind: string; filePath: string; startLine?: number }>();
       const seenEdges = new Set<string>();
       let edgeCount = 0;
 
-      for (const match of matches) {
-        const exactMatch = match.node.name === symbol || match.node.name.endsWith(`.${symbol}`) || match.node.name.endsWith(`::${symbol}`);
-        if (!exactMatch && matches.length > 1) continue;
-        const impact = cg.getImpactRadius(match.node.id, depth);
+      for (const target of targets) {
+        const impact = cg.getImpactRadius(target.id, depth);
         for (const [id, n] of impact.nodes) {
           mergedNodes.set(id, { name: n.name, kind: n.kind, filePath: n.filePath, startLine: n.startLine });
         }
@@ -2142,19 +2188,15 @@ program
         }
       }
 
-      // Fallback to top match if exact filter removed everything
-      if (mergedNodes.size === 0 && matches[0]) {
-        const impact = cg.getImpactRadius(matches[0].node.id, depth);
-        for (const [id, n] of impact.nodes) {
-          mergedNodes.set(id, { name: n.name, kind: n.kind, filePath: n.filePath, startLine: n.startLine });
-        }
-        edgeCount = impact.edges.length;
-      }
-
       if (options.json) {
         console.log(JSON.stringify({
           symbol,
           depth,
+          targets: targets.map((t) => ({
+            qualifiedName: t.qualifiedName, kind: t.kind, language: t.language,
+            filePath: t.filePath, startLine: t.startLine,
+          })),
+          ambiguous,
           nodeCount: mergedNodes.size,
           edgeCount,
           affected: Array.from(mergedNodes.values()),
@@ -2180,6 +2222,7 @@ program
           }
           console.log();
         }
+        printAmbiguityNote(symbol, targets, ambiguous);
       }
 
       cg.destroy();
diff --git a/src/graph/symbol-lookup.ts b/src/graph/symbol-lookup.ts
new file mode 100644
index 000000000..afeefed38
--- /dev/null
+++ b/src/graph/symbol-lookup.ts
@@ -0,0 +1,178 @@
+/**
+ * Symbol Lookup — the single "what did the user mean by this name?" path.
+ *
+ * Every verb that takes a symbol NAME from a human (or an agent) has to turn
+ * that string into node(s). `codegraph_node` and `codegraph_explore` went
+ * through the matcher below; the `callers` / `callees` / `impact` CLI verbs
+ * carried their own ad-hoc filter instead:
+ *
+ *     node.name === symbol || node.name.endsWith('.' + symbol)
+ *
+ * which compares the query against the BARE name only. That produced two
+ * opposite failures in the same repository:
+ *
+ *   - a bare name over-reported: `callers group` silently merged the callers of
+ *     every distinct symbol named `group` — in any language — into one list
+ *     headed "Callers of group", with nothing saying they were different
+ *     symbols;
+ *   - a qualified name under-reported: `Foo.Bar.baz` can never equal a bare
+ *     `baz`, so every candidate failed the filter and the code fell through to
+ *     an arbitrary top-of-FTS hit — or reported "not found" for a symbol that
+ *     plainly exists.
+ *
+ * Both are fixed by routing all of them through one resolver, which this module
+ * owns so the CLI and the MCP tools cannot drift apart again.
+ */
+
+import type { Node } from '../types';
+
+/** Rust path prefixes that name no directory (`crate::x`, `super::y`). */
+const RUST_PATH_PREFIXES = new Set(['crate', 'super', 'self']);
+
+/** Does this query carry any scope qualifier at all? */
+export function isQualifiedSymbol(symbol: string): boolean {
+  return /[.\/]|::/.test(symbol);
+}
+
+/** The bare identifier at the end of a qualified query (arity spelling stripped). */
+export function lastQualifierPart(symbol: string): string {
+  const noArity = symbol.replace(/\/\d{1,3}$/, '') || symbol;
+  const parts = noArity.split(/::|[./]/).filter((p) => p.length > 0);
+  return parts[parts.length - 1] ?? symbol;
+}
+
+/**
+ * Rewrite every scope separator to `.` so a query and a stored qualifiedName
+ * written in different conventions can be compared directly. The extractors
+ * join hierarchy with `::` while users type the language's own spelling
+ * (`Session.request`, `stage_apply::run`, `pkg/mod.Fn`).
+ */
+function canonicalScope(text: string): string {
+  return text.replace(/::/g, '.').replace(/\//g, '.');
+}
+
+/**
+ * Does `node` satisfy the user's symbol query?
+ *
+ * Bare queries match the name. Qualified queries are checked against the
+ * qualifiedName under both separator conventions, then — for languages whose
+ * hierarchy lives in the file path rather than the name (Rust modules, Python
+ * packages) — against the path.
+ */
+export function matchesSymbol(node: Node, symbol: string): boolean {
+  // Erlang arity spelling (`fn/3`, `mod:fn/3`): when the node's qualifiedName
+  // carries an arity (#1610) the written arity must match exactly, and the rest
+  // of the comparison runs on the arity-less spelling. A node with no arity
+  // keeps the original symbol (a `/` there means a path-ish name instead).
+  const aritySpelling = /^(.+)\/(\d{1,3})$/.exec(symbol);
+  if (aritySpelling) {
+    const nodeArity = /\/(\d{1,3})$/.exec(node.qualifiedName ?? '')?.[1];
+    if (nodeArity !== undefined) {
+      if (nodeArity !== aritySpelling[2]) return false;
+      symbol = aritySpelling[1]!;
+    }
+  }
+
+  if (node.name === symbol) return true;
+  // File basename match ("product-card" matches "product-card.liquid").
+  if (node.kind === 'file' && node.name.replace(/\.[^.]+$/, '') === symbol) return true;
+
+  if (!isQualifiedSymbol(symbol)) return false;
+  const parts = symbol.split(/::|[./]/).filter((p) => p.length > 0);
+  if (parts.length < 2) return false;
+
+  const lastPart = parts[parts.length - 1]!;
+  if (node.name !== lastPart) return false;
+
+  // Stage 1: qualified-name containment under the extractor's `::` convention.
+  if (node.qualifiedName.includes(parts.join('::'))) return true;
+
+  // Stage 1b: boundary-aligned suffix under a canonical separator.
+  //
+  // Splitting on EVERY separator assumes no scope component contains one —
+  // false for any language whose module names are themselves dotted (Elixir
+  // `AppWeb.Format`, a Java/C# package, a Python dotted module). There the
+  // stored qualifiedName is `AppWeb.Format::group`, so the stage-1 spelling
+  // `AppWeb::Format::group` cannot match and a perfectly precise query
+  // resolved to nothing. Canonicalising both sides and requiring the match to
+  // land on a separator boundary handles both conventions with one rule, and
+  // is strictly tighter than the `includes` above.
+  const canonicalQuery = canonicalScope(symbol);
+  const canonicalNode = canonicalScope(node.qualifiedName);
+  if (canonicalNode === canonicalQuery || canonicalNode.endsWith(`.${canonicalQuery}`)) {
+    return true;
+  }
+
+  // Stage 2: file-path containment. Rust modules and Python packages are not in
+  // qualifiedName — they are encoded in the path — so `stage_apply::run`
+  // matches a `run` in any file with a `stage_apply` path segment.
+  const containerHints = parts.slice(0, -1).filter((p) => !RUST_PATH_PREFIXES.has(p));
+  if (containerHints.length === 0) return false;
+  const segments = node.filePath.split('/').filter((s) => s.length > 0);
+  return containerHints.every((hint) =>
+    segments.some((seg) => seg === hint || seg.replace(/\.[^.]+$/, '') === hint)
+  );
+}
+
+/** The slice of CodeGraph a symbol lookup needs — keeps this module testable. */
+export interface SymbolLookupHost {
+  getNodesByName(name: string): Node[];
+  searchNodes(query: string, options?: { limit?: number }): Array<{ node: Node }>;
+  generatedFilePredicate(paths: string[]): (path: string) => boolean;
+}
+
+export interface SymbolLookupResult {
+  /** Every definition the query names, keepers before generated stubs. */
+  nodes: Node[];
+  /**
+   * The query named more than one distinct definition. Callers that aggregate
+   * across all of them MUST surface this — an aggregate presented as one
+   * symbol's answer is the over-reporting failure described at the top.
+   */
+  ambiguous: boolean;
+}
+
+/**
+ * Resolve a user-supplied symbol name to the definitions it names.
+ *
+ * The exact-name index is consulted FIRST and is authoritative: it is complete
+ * and uncapped, whereas FTS ranks and truncates, and tokenises away `::` — so
+ * a qualified query could miss a symbol that exists, or land on whatever
+ * happened to rank first. FTS remains as the fallback for the fuzzy cases it is
+ * genuinely good at (file basenames, partial names).
+ */
+export function lookupSymbolNodes(cg: SymbolLookupHost, symbol: string): SymbolLookupResult {
+  const qualified = isQualifiedSymbol(symbol);
+
+  // Exact-name index, then filter by the qualifier the user actually wrote.
+  const tail = qualified ? lastQualifierPart(symbol) : symbol;
+  let nodes = tail ? cg.getNodesByName(tail) : [];
+  if (qualified) nodes = nodes.filter((n) => matchesSymbol(n, symbol));
+
+  if (nodes.length === 0) {
+    const hits = cg.searchNodes(symbol, { limit: 50 }).map((h) => h.node);
+    const exact = hits.filter((n) => matchesSymbol(n, symbol));
+    if (exact.length > 0) {
+      nodes = exact;
+    } else if (!qualified && hits[0]) {
+      // A bare name with no exact definition may still mean a file basename.
+      nodes = [hits[0]];
+    }
+    // A qualified query with no exact match resolves to NOTHING rather than a
+    // misleading fuzzy hit (#173).
+  }
+
+  if (nodes.length === 0) return { nodes: [], ambiguous: false };
+
+  // Keepers before generated stubs (.pb.go and friends), stable otherwise.
+  const isGenerated = cg.generatedFilePredicate(nodes.map((n) => n.filePath));
+  const ranked = [...nodes].sort(
+    (a, b) => (isGenerated(a.filePath) ? 1 : 0) - (isGenerated(b.filePath) ? 1 : 0)
+  );
+  return { nodes: ranked, ambiguous: ranked.length > 1 };
+}
+
+/** One-line "kind at path:line" label used when disclosing an ambiguous query. */
+export function describeSymbolNode(node: Node): string {
+  return `${node.kind} ${node.qualifiedName || node.name} (${node.language}) — ${node.filePath}:${node.startLine}`;
+}
diff --git a/src/mcp/tools.ts b/src/mcp/tools.ts
index 5c23f675d..535490067 100644
--- a/src/mcp/tools.ts
+++ b/src/mcp/tools.ts
@@ -32,6 +32,7 @@ import {
 import type { PendingFile } from '../sync';
 import type { Node, Edge, SearchResult, Subgraph, NodeKind } from '../types';
 import { isTestFile, normalizeNameToken } from '../search/query-utils';
+import { lastQualifierPart, matchesSymbol as matchesSymbolShared } from '../graph/symbol-lookup';
 import { extractQueryPaths, queryMightContainPaths } from '../search/query-paths';
 import {
   existsSync,
@@ -104,13 +105,9 @@ const MAX_INPUT_LENGTH = 10_000;
 const MAX_PATH_LENGTH = 4_096;
 
 /**
- * Rust path roots that have no file-system equivalent — `crate` is the
- * current crate, `super` is the parent module, `self` is the current
- * module. Used by `matchesSymbol` to strip these before file-path
- * matching so `crate::configurator::stage_apply::run` resolves the
- * same as `configurator::stage_apply::run`.
+ * (Rust path roots — `crate`/`super`/`self` — moved to
+ * ../graph/symbol-lookup along with `matchesSymbol`.)
  */
-const RUST_PATH_PREFIXES = new Set(['crate', 'super', 'self']);
 
 /**
  * Node kinds that contain other symbols. For these, `codegraph_node` with
@@ -123,15 +120,10 @@ const CONTAINER_NODE_KINDS = new Set<NodeKind>([
 ]);
 
 /**
- * Last `::` / `.` / `/`-separated segment of a qualified symbol. An Erlang
- * arity tail (`mod::fn/3`, `fn/3`) is stripped first — the useful last segment
- * is the function name, never the digits (#1610).
+ * Symbol-name resolution lives in ../graph/symbol-lookup so the CLI verbs and
+ * these tools share one path. `lastQualifierPart` and `matchesSymbol` are
+ * re-exported through this module's own call sites unchanged.
  */
-function lastQualifierPart(symbol: string): string {
-  const noArity = symbol.replace(/\/\d{1,3}$/, '') || symbol;
-  const parts = noArity.split(/::|[./]/).filter((p) => p.length > 0);
-  return parts[parts.length - 1] ?? symbol;
-}
 
 /**
  * Normalize Erlang-native symbol spellings in an explore query into the shapes
@@ -6728,53 +6720,7 @@ export class ToolHandler {
    *      Python — `stage_apply::run` matches a `run` in `stage_apply.rs`)
    */
   private matchesSymbol(node: Node, symbol: string): boolean {
-    // Erlang arity spelling (`fn/3`, `mod:fn/3` → normalized `mod.fn/3`): when
-    // the node's qualifiedName carries an arity (`mod::fn/3`, #1610), the
-    // written arity must match it exactly; the remaining comparison then runs
-    // on the arity-less spelling. A node with no arity in its qualifiedName
-    // keeps the original symbol (a `/` there means a path-ish name instead).
-    const aritySpelling = /^(.+)\/(\d{1,3})$/.exec(symbol);
-    if (aritySpelling) {
-      const nodeArity = /\/(\d{1,3})$/.exec(node.qualifiedName ?? '')?.[1];
-      if (nodeArity !== undefined) {
-        if (nodeArity !== aritySpelling[2]) return false;
-        symbol = aritySpelling[1]!;
-      }
-    }
-    // Simple name match
-    if (node.name === symbol) return true;
-    // File basename match (e.g., "product-card" matches "product-card.liquid")
-    if (node.kind === 'file' && node.name.replace(/\.[^.]+$/, '') === symbol) return true;
-
-    // Qualified-name lookups: split on any supported separator. `\w` keeps
-    // identifier chars (incl. `_`) intact; everything else is treated as
-    // a separator we tolerate.
-    if (!/[.\/]|::/.test(symbol)) return false;
-    const parts = symbol.split(/::|[./]/).filter((p) => p.length > 0);
-    if (parts.length < 2) return false;
-
-    const lastPart = parts[parts.length - 1]!;
-    if (node.name !== lastPart) return false;
-
-    // Stage 1: qualified-name suffix match. The extractor joins the
-    // semantic hierarchy with `::`, so `Session.request` and
-    // `Session::request` both become `Session::request` here.
-    const colonSuffix = parts.join('::');
-    if (node.qualifiedName.includes(colonSuffix)) return true;
-
-    // Stage 2: file-path containment. Rust modules and Python packages
-    // are not in `qualifiedName` — they're encoded in the file path. So
-    // `stage_apply::run` matches a `run` in any file whose path
-    // contains a `stage_apply` segment (with or without an extension).
-    //
-    // Filter out Rust path prefixes that have no file-system equivalent.
-    const containerHints = parts.slice(0, -1).filter((p) => !RUST_PATH_PREFIXES.has(p));
-    if (containerHints.length === 0) return false;
-
-    const segments = node.filePath.split('/').filter((s) => s.length > 0);
-    return containerHints.every((hint) =>
-      segments.some((seg) => seg === hint || seg.replace(/\.[^.]+$/, '') === hint)
-    );
+    return matchesSymbolShared(node, symbol);
   }
 
   /**