diff --git a/gitnexus/src/mcp/output-budget.ts b/gitnexus/src/mcp/output-budget.ts
new file mode 100644
index 0000000000..032d3cd413
--- /dev/null
+++ b/gitnexus/src/mcp/output-budget.ts
@@ -0,0 +1,58 @@
+const BUDGETED_TOOLS = new Set(['query', 'context', 'impact']);
+
+export const MCP_TOKEN_ESTIMATE_BYTES = 4;
+export const MCP_TRUNCATION_MARKER = '\n…';
+
+function parsePositiveInteger(value: unknown, source: string): number {
+  if (typeof value === 'number' && Number.isSafeInteger(value) && value > 0) return value;
+  if (typeof value === 'string' && /^[1-9]\d*$/.test(value.trim())) {
+    const parsed = Number(value.trim());
+    if (Number.isSafeInteger(parsed)) return parsed;
+  }
+  throw new Error(`${source} must be a positive integer.`);
+}
+
+export function resolveMcpMaxTokens(
+  toolName: string,
+  args: Record<string, unknown> | undefined,
+  env: NodeJS.ProcessEnv = process.env,
+): number | undefined {
+  if (!BUDGETED_TOOLS.has(toolName)) return undefined;
+  if (args?.maxTokens !== undefined) return parsePositiveInteger(args.maxTokens, 'maxTokens');
+
+  const configured = env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+  if (configured === undefined || configured.trim() === '') return undefined;
+  return parsePositiveInteger(configured, 'GITNEXUS_MCP_DEFAULT_MAX_TOKENS');
+}
+
+function utf8Prefix(text: string, maxBytes: number): string {
+  let bytes = 0;
+  const codePoints: string[] = [];
+  for (const codePoint of text) {
+    const codePointBytes = Buffer.byteLength(codePoint, 'utf8');
+    if (bytes + codePointBytes > maxBytes) break;
+    codePoints.push(codePoint);
+    bytes += codePointBytes;
+  }
+  return codePoints.join('');
+}
+
+export function applyMcpMaxTokens(text: string, maxTokens: number | undefined): string {
+  if (maxTokens === undefined) return text;
+
+  const textBytes = Buffer.byteLength(text, 'utf8');
+  if (maxTokens >= Math.ceil(textBytes / MCP_TOKEN_ESTIMATE_BYTES)) return text;
+
+  const maxBytes = maxTokens * MCP_TOKEN_ESTIMATE_BYTES;
+  const markerBytes = Buffer.byteLength(MCP_TRUNCATION_MARKER, 'utf8');
+  return utf8Prefix(text, Math.max(0, maxBytes - markerBytes)) + MCP_TRUNCATION_MARKER;
+}
+
+export function withoutMcpBudgetArg(
+  args: Record<string, unknown> | undefined,
+): Record<string, unknown> | undefined {
+  if (!args || !Object.prototype.hasOwnProperty.call(args, 'maxTokens')) return args;
+  const backendArgs = { ...args };
+  delete backendArgs.maxTokens;
+  return backendArgs;
+}
diff --git a/gitnexus/src/mcp/server.ts b/gitnexus/src/mcp/server.ts
index d4a7c58aa8..bd8b08d03b 100644
--- a/gitnexus/src/mcp/server.ts
+++ b/gitnexus/src/mcp/server.ts
@@ -27,6 +27,7 @@ import { GITNEXUS_TOOLS } from './tools.js';
 import { installGlobalStdoutSentinel } from './stdio-context.js';
 import type { LocalBackend } from './local/local-backend.js';
 import { getResourceDefinitions, getResourceTemplates, readResource } from './resources.js';
+import { applyMcpMaxTokens, resolveMcpMaxTokens, withoutMcpBudgetArg } from './output-budget.js';
 
 /**
  * Next-step hints appended to tool responses.
@@ -165,9 +166,12 @@ export function createMCPServer(backend: LocalBackend): Server {
   // Handle tool calls — append next-step hints to guide agent workflow
   server.setRequestHandler(CallToolRequestSchema, async (request) => {
     const { name, arguments: args } = request.params;
+    let maxTokens: number | undefined;
 
     try {
-      const result = await backend.callTool(name, args);
+      const typedArgs = args as Record<string, unknown> | undefined;
+      maxTokens = resolveMcpMaxTokens(name, typedArgs);
+      const result = await backend.callTool(name, withoutMcpBudgetArg(typedArgs));
       const resultText = typeof result === 'string' ? result : JSON.stringify(result, null, 2);
       const hint = getNextStepHint(name, args as Record<string, any> | undefined);
 
@@ -175,7 +179,7 @@ export function createMCPServer(backend: LocalBackend): Server {
         content: [
           {
             type: 'text',
-            text: resultText + hint,
+            text: applyMcpMaxTokens(resultText + hint, maxTokens),
           },
         ],
       };
@@ -185,7 +189,7 @@ export function createMCPServer(backend: LocalBackend): Server {
         content: [
           {
             type: 'text',
-            text: `Error: ${message}`,
+            text: applyMcpMaxTokens(`Error: ${message}`, maxTokens),
           },
         ],
         isError: true,
diff --git a/gitnexus/src/mcp/tools.ts b/gitnexus/src/mcp/tools.ts
index 57f71f6752..ec9a96ac57 100644
--- a/gitnexus/src/mcp/tools.ts
+++ b/gitnexus/src/mcp/tools.ts
@@ -182,6 +182,12 @@ SERVICE: optional monorepo path prefix (POSIX-style, case-sensitive segments). W
           description: 'Include full symbol source code (default: false)',
           default: false,
         },
+        maxTokens: {
+          type: 'integer',
+          minimum: 1,
+          description:
+            'Maximum estimated tokens in the complete formatted MCP response. Explicit request overrides GITNEXUS_MCP_DEFAULT_MAX_TOKENS.',
+        },
         repo: {
           type: 'string',
           description:
@@ -311,6 +317,12 @@ SERVICE: optional monorepo path prefix (case-sensitive path segments). When "rep
           description: 'Include full symbol source code (default: false)',
           default: false,
         },
+        maxTokens: {
+          type: 'integer',
+          minimum: 1,
+          description:
+            'Maximum estimated tokens in the complete formatted MCP response. Explicit request overrides GITNEXUS_MCP_DEFAULT_MAX_TOKENS.',
+        },
         repo: {
           type: 'string',
           description:
@@ -576,6 +588,12 @@ SERVICE: optional monorepo path prefix (case-sensitive path segments). When "rep
             'When true, returns target, summary, risk, byDepthCounts, affected_processes, and affected_modules — omits byDepth. Single-repo only; ignored in group mode (@groupName). Use for hub symbols to get actionable signal without output explosion.',
           default: false,
         },
+        maxTokens: {
+          type: 'integer',
+          minimum: 1,
+          description:
+            'Maximum estimated tokens in the complete formatted MCP response. Explicit request overrides GITNEXUS_MCP_DEFAULT_MAX_TOKENS.',
+        },
         timeoutMs: {
           type: 'number',
           description:
diff --git a/gitnexus/test/unit/mcp-output-budget.test.ts b/gitnexus/test/unit/mcp-output-budget.test.ts
new file mode 100644
index 0000000000..8e568f0149
--- /dev/null
+++ b/gitnexus/test/unit/mcp-output-budget.test.ts
@@ -0,0 +1,61 @@
+import { describe, expect, it } from 'vitest';
+import {
+  applyMcpMaxTokens,
+  MCP_TRUNCATION_MARKER,
+  resolveMcpMaxTokens,
+  withoutMcpBudgetArg,
+} from '../../src/mcp/output-budget.js';
+
+describe('MCP output budget helpers', () => {
+  it('returns the original string byte-for-byte without a configured budget', () => {
+    const text = 'alpha😀omega';
+    expect(applyMcpMaxTokens(text, undefined)).toBe(text);
+  });
+
+  it('uses the complete marker and stays within a one-token budget', () => {
+    const text = applyMcpMaxTokens('this response is too long', 1);
+    expect(text).toBe(MCP_TRUNCATION_MARKER);
+    expect(Buffer.byteLength(text, 'utf8')).toBe(4);
+  });
+
+  it('never splits a multi-byte Unicode code point', () => {
+    const text = applyMcpMaxTokens('😀😀😀😀', 3);
+    expect(text.endsWith(MCP_TRUNCATION_MARKER)).toBe(true);
+    expect(text).not.toContain('\uFFFD');
+    expect(Buffer.byteLength(text, 'utf8')).toBeLessThanOrEqual(12);
+  });
+
+  it('rejects malformed environment defaults for budgeted tools', () => {
+    expect(() =>
+      resolveMcpMaxTokens('query', undefined, {
+        GITNEXUS_MCP_DEFAULT_MAX_TOKENS: '1.5',
+      }),
+    ).toThrow(/positive integer/i);
+  });
+
+  it('lets a valid explicit value override a malformed environment default', () => {
+    expect(
+      resolveMcpMaxTokens(
+        'impact',
+        { maxTokens: 17 },
+        {
+          GITNEXUS_MCP_DEFAULT_MAX_TOKENS: 'invalid',
+        },
+      ),
+    ).toBe(17);
+  });
+
+  it('ignores the environment default for tools without output budgets', () => {
+    expect(
+      resolveMcpMaxTokens('cypher', undefined, {
+        GITNEXUS_MCP_DEFAULT_MAX_TOKENS: 'invalid',
+      }),
+    ).toBeUndefined();
+  });
+
+  it('removes only the transport-level maxTokens argument', () => {
+    const args = { search_query: 'auth', maxTokens: 20, repo: 'app' };
+    expect(withoutMcpBudgetArg(args)).toEqual({ search_query: 'auth', repo: 'app' });
+    expect(args).toEqual({ search_query: 'auth', maxTokens: 20, repo: 'app' });
+  });
+});
diff --git a/gitnexus/test/unit/server.test.ts b/gitnexus/test/unit/server.test.ts
index f2ac995da6..84850ffa41 100644
--- a/gitnexus/test/unit/server.test.ts
+++ b/gitnexus/test/unit/server.test.ts
@@ -43,6 +43,27 @@ function createMockBackend(overrides: Record<string, any> = {}): any {
   };
 }
 
+async function callToolThroughServer(
+  backend: ReturnType<typeof createMockBackend>,
+  name: string,
+  args: Record<string, unknown>,
+): Promise<{ text: string; isError: boolean }> {
+  const server = createMCPServer(backend);
+  const client = new Client({ name: 'budget-test-client', version: '0.0.0' });
+  const [clientTransport, serverTransport] = InMemoryTransport.createLinkedPair();
+
+  try {
+    await Promise.all([server.connect(serverTransport), client.connect(clientTransport)]);
+    const response = await client.callTool({ name, arguments: args });
+    const text = response.content.find((item) => item.type === 'text')?.text;
+    if (typeof text !== 'string') throw new Error('Expected an MCP text response');
+    return { text, isError: response.isError === true };
+  } finally {
+    await client.close();
+    await server.close();
+  }
+}
+
 // ─── createMCPServer ─────────────────────────────────────────────────
 
 describe('createMCPServer', () => {
@@ -105,6 +126,107 @@ describe('getNextStepHint (via tool call response)', () => {
   });
 });
 
+describe('MCP output budgets', () => {
+  it('leaves the complete formatted response unchanged when no budget is configured', async () => {
+    const previous = process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+    delete process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+    try {
+      const backend = createMockBackend({
+        callTool: vi.fn().mockResolvedValue({ payload: 'complete' }),
+      });
+      const { text, isError } = await callToolThroughServer(backend, 'query', {
+        search_query: 'auth',
+      });
+      expect(isError).toBe(false);
+      expect(text).toContain('"payload": "complete"');
+      expect(text).toContain('**Next:**');
+      expect(text.endsWith('\n…')).toBe(false);
+    } finally {
+      if (previous === undefined) delete process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+      else process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS = previous;
+    }
+  });
+
+  it('applies explicit maxTokens to the complete response deterministically and UTF-8 safely', async () => {
+    const backend = createMockBackend({
+      callTool: vi.fn().mockResolvedValue({ payload: '😀'.repeat(100) }),
+    });
+    const args = { search_query: 'auth', maxTokens: 8 };
+
+    const first = await callToolThroughServer(backend, 'query', args);
+    const second = await callToolThroughServer(backend, 'query', args);
+
+    expect(first.isError).toBe(false);
+    expect(first.text).toBe(second.text);
+    expect(Buffer.byteLength(first.text, 'utf8')).toBeLessThanOrEqual(8 * 4);
+    expect(first.text.endsWith('\n…')).toBe(true);
+    expect(first.text).not.toContain('\uFFFD');
+    expect(backend.callTool).toHaveBeenCalledWith('query', { search_query: 'auth' });
+  });
+
+  it('uses the environment default when maxTokens is omitted', async () => {
+    const previous = process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+    process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS = '8';
+    try {
+      const backend = createMockBackend({
+        callTool: vi.fn().mockResolvedValue({ payload: 'x'.repeat(200) }),
+      });
+      const { text } = await callToolThroughServer(backend, 'context', { name: 'auth' });
+      expect(Buffer.byteLength(text, 'utf8')).toBeLessThanOrEqual(8 * 4);
+      expect(text.endsWith('\n…')).toBe(true);
+    } finally {
+      if (previous === undefined) delete process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+      else process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS = previous;
+    }
+  });
+
+  it('lets an explicit request override the environment default', async () => {
+    const previous = process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+    process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS = '1';
+    try {
+      const backend = createMockBackend({
+        callTool: vi.fn().mockResolvedValue({ payload: 'complete' }),
+      });
+      const { text } = await callToolThroughServer(backend, 'impact', {
+        target: 'auth',
+        direction: 'upstream',
+        maxTokens: 200,
+      });
+      expect(text).toContain('"payload": "complete"');
+      expect(text).toContain('**Next:**');
+      expect(text.endsWith('\n…')).toBe(false);
+    } finally {
+      if (previous === undefined) delete process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS;
+      else process.env.GITNEXUS_MCP_DEFAULT_MAX_TOKENS = previous;
+    }
+  });
+
+  it('rejects a non-positive explicit maxTokens before backend execution', async () => {
+    const backend = createMockBackend();
+    const { text, isError } = await callToolThroughServer(backend, 'query', {
+      search_query: 'auth',
+      maxTokens: 0,
+    });
+    expect(isError).toBe(true);
+    expect(text).toMatch(/maxTokens.*positive integer/i);
+    expect(backend.callTool).not.toHaveBeenCalled();
+  });
+
+  it('applies a valid budget to backend error text', async () => {
+    const backend = createMockBackend({
+      callTool: vi.fn().mockRejectedValue(new Error('😀'.repeat(100))),
+    });
+    const { text, isError } = await callToolThroughServer(backend, 'context', {
+      name: 'auth',
+      maxTokens: 8,
+    });
+    expect(isError).toBe(true);
+    expect(Buffer.byteLength(text, 'utf8')).toBeLessThanOrEqual(8 * 4);
+    expect(text.endsWith('\n…')).toBe(true);
+    expect(text).not.toContain('\uFFFD');
+  });
+});
+
 // ─── Tool handler error handling ──────────────────────────────────────
 
 describe('server error handling', () => {
diff --git a/gitnexus/test/unit/tools.test.ts b/gitnexus/test/unit/tools.test.ts
index 4757fd9f7c..d55a214901 100644
--- a/gitnexus/test/unit/tools.test.ts
+++ b/gitnexus/test/unit/tools.test.ts
@@ -185,6 +185,16 @@ describe('GITNEXUS_TOOLS', () => {
     expect(impactTool.description).toContain('truncatedBy');
   });
 
+  it.each(['query', 'context', 'impact'])(
+    '%s advertises an optional positive maxTokens budget',
+    (name) => {
+      const tool = GITNEXUS_TOOLS.find((definition) => definition.name === name)!;
+      const maxTokens = tool.inputSchema.properties.maxTokens;
+      expect(maxTokens).toMatchObject({ type: 'integer', minimum: 1 });
+      expect(tool.inputSchema.required).not.toContain('maxTokens');
+    },
+  );
+
   it('rename tool requires new_name', () => {
     const renameTool = GITNEXUS_TOOLS.find((t) => t.name === 'rename')!;
     expect(renameTool.inputSchema.required).toContain('new_name');