LLM response contains markdown code fences that break JSON.parse()
Problem
When prompting an LLM to return JSON, the model often wraps the output in markdown code fences (```json ... ```) even when explicitly told not to. Passing this raw response to JSON.parse() throws SyntaxError. This happens with GPT-4o, Claude, and Gemini — any model that has been fine-tuned to be "helpful" tends to add markdown formatting.
Error Output
SyntaxError: Unexpected token '`' at JSON.parse (<anonymous>)
Unverified for your environment
Select your OS to check compatibility.
1 Fix
Strip markdown code fences before JSON.parse(), or use response_format: json_object
LLMs are trained to be helpful by formatting code in markdown. Even with instructions to return raw JSON, models frequently wrap output in ```json ... ```. The fix is to strip fences defensively or use structured output modes.
Trust Score
6 verifications
- 1
Strip code fences before parsing
Add a utility function to clean LLM output:
typescriptfunction extractJSON(text: string): any { // Remove markdown code fences: ```json ... ``` or ``` ... ``` const cleaned = text .replace(/^```(?:json)?s*/m, '') .replace(/s*```s*$/m, '') .trim() return JSON.parse(cleaned) } - 2
Use response_format for OpenAI (GPT-4o)
Force JSON output mode:
typescriptconst response = await openai.chat.completions.create({ model: 'gpt-4o', messages: [...], response_format: { type: 'json_object' }, }) - 3
For Claude: use tool use to get structured output
Define a tool with the JSON schema you want. Claude's tool_input is always valid JSON:
typescriptconst response = await anthropic.messages.create({ model: 'claude-opus-4-6', tools: [{ name: 'output', input_schema: { type: 'object', properties: {...} } }], tool_choice: { type: 'tool', name: 'output' }, messages: [...], }) const result = response.content.find(b => b.type === 'tool_use')?.input
Validation
JSON.parse() no longer throws. Response parsing works correctly for all LLM responses.
Verification Summary
Sign in to verify this fix
Environment
- Product
- OpenAI API / Claude API
- Environment
- production
Submitted by
Alex Chen
2450 rep