MCP SDK

@lookspan/mcp is Lookspan’s MCP-native SDK. It wraps any McpClient and emits a tool_call span per MCP tool invocation, without changing your agent code. It’s also the foundation the OpenAI and Anthropic observers are built on, so it re-exports the shared HttpSpanExporter and span types.

Install

npm install @lookspan/mcp

Usage

import { wrapMcpClient, HttpSpanExporter } from '@lookspan/mcp';

const exporter = new HttpSpanExporter({ endpoint: 'http://127.0.0.1:3100/api/ingest' });
const { client } = wrapMcpClient(mcpClient, { exporter, agentId: 'my-agent' });

// Use it exactly as before — every callTool emits a tool_call span.
await client.callTool({ name: 'read_file', arguments: { path: '/tmp/foo.txt' } });
await exporter.flush();

wrapMcpClient returns { client, traceId, flush }: the wrapped client, the trace id all its tool calls share, and a flush() helper.

wrapMcpClient options

Option	Type	Default	What it does
exporter	SpanExporter	(required)	Where spans are sent (e.g. an HttpSpanExporter).
agentId	string	—	Attribution label for the dashboard.
sessionId	string	—	Group related agents into a session.
traceId	string	auto	Reuse a trace id (e.g. to join an existing trace).
rootSpanId	string	—	Parent span for the tool calls (nests them under a step).
attributes	Record<string, unknown>	—	Extra attributes stamped on every span.

Each tool call becomes a span named mcp.tool.<toolName> of type tool_call, with the call arguments captured as input and the result as output.

HttpSpanExporter options

Option	Type	Default	What it does
endpoint	string	(required)	Ingest URL.
flushIntervalMs	number	—	Auto-flush interval.
maxBatchSize	number	—	Spans per POST before flushing.
maxRetries	number	2	Extra POST attempts on transient failure.
retryBackoffMs	number	200	Base backoff, doubled each retry.
source	string	—	Label for the exporter source.

Spans are buffered and sent in batches; call flush() before your process exits to drain anything still buffered.