Import Formats¶

Supported log formats for importing traces into Assay.

Overview¶

Assay can import agent sessions from various sources:

MCP Inspector¶

The primary format for MCP-based agents.

Export from MCP Inspector¶

1. Run your agent session in MCP Inspector
2. File → Export Session → JSON
3. Save as session.json

Import¶

assay import --format mcp-inspector session.json

Format Structure¶

{
  "messages": [
    {
      "jsonrpc": "2.0",
      "id": 1,
      "method": "tools/call",
      "params": {
        "name": "get_customer",
        "arguments": { "id": "cust_123" }
      }
    },
    {
      "jsonrpc": "2.0",
      "id": 1,
      "result": {
        "content": [
          { "type": "text", "text": "{\"name\": \"Alice\"}" }
        ],
        "isError": false
      }
    }
  ]
}

Field Mapping¶

JSON-RPC 2.0¶

Raw JSON-RPC messages, useful for custom MCP implementations.

Import¶

assay import --format jsonrpc messages.json

Format Structure¶

[
  {
    "jsonrpc": "2.0",
    "id": "call_001",
    "method": "tools/call",
    "params": {
      "name": "apply_discount",
      "arguments": { "percent": 25 }
    }
  },
  {
    "jsonrpc": "2.0",
    "id": "call_001",
    "result": { "success": true }
  }
]

Notes¶

• Array of messages (not wrapped in messages object)
• id field links requests to responses
• Only tools/call method is processed

LangChain (Coming Soon)¶

Import from LangChain's tracing format.

Expected Usage¶

assay import --format langchain langchain_trace.json

Format (Preview)¶

{
  "runs": [
    {
      "id": "run_abc123",
      "name": "Tool",
      "inputs": {
        "tool": "get_customer",
        "tool_input": { "id": "cust_123" }
      },
      "outputs": {
        "output": { "name": "Alice" }
      }
    }
  ]
}

Status¶

Currently in development. Track progress at GitHub Issue #42.

LlamaIndex (Coming Soon)¶

Import from LlamaIndex's instrumentation.

Expected Usage¶

assay import --format llamaindex llamaindex_trace.json

Status¶

Planned for v1.1. Track progress at GitHub Issue #43.

Output Format¶

All imports produce Assay's normalized trace format:

{"type":"tool_call","id":"1","tool":"get_customer","arguments":{"id":"cust_123"},"timestamp":"2025-12-27T10:00:00Z"}
{"type":"tool_result","id":"1","result":{"name":"Alice"},"timestamp":"2025-12-27T10:00:01Z"}

Fields¶

Custom Formats¶

For unsupported formats, convert to JSON-RPC manually:

import json

def convert_custom_to_jsonrpc(custom_log):
    messages = []
    for i, entry in enumerate(custom_log):
        # Request
        messages.append({
            "jsonrpc": "2.0",
            "id": str(i),
            "method": "tools/call",
            "params": {
                "name": entry["tool_name"],
                "arguments": entry["inputs"]
            }
        })
        # Response
        messages.append({
            "jsonrpc": "2.0",
            "id": str(i),
            "result": entry["outputs"]
        })
    return messages

# Save and import
with open("converted.json", "w") as f:
    json.dump(convert_custom_to_jsonrpc(my_log), f)

Then import:

assay import --format jsonrpc converted.json

Troubleshooting¶

"No tool calls found"¶

The session might not contain tools/call messages:

# Check what methods are in the file
cat session.json | jq '.messages[].method' | sort | uniq

"Invalid JSON"¶

Validate the file:

jq . session.json > /dev/null
# If no output, JSON is valid
# If error, fix the syntax

"Missing required field"¶

Check that each call has: - params.name (tool name) - params.arguments (can be empty {})

See Also¶

• assay import
• Traces
• MCP Quick Start