mcp-parser

Snapshot, parse, validate, and document Model Context Protocol servers.

MCP Protocol Compatibility

snapshot requests protocol revision 2025-11-25 during initialize, sends it as the MCP-Protocol-Version header on HTTP transports, and records whatever revision the server answers with as mcpVersion. Servers that negotiate an older revision are snapshotted at the revision they return.

Parsing, validation, and generation are revision-independent: they read an mcp.json document, whatever revision produced it.

Transports:

Transport	Support
stdio	Full
Streamable HTTP	Full, including server-minted sessions (Mcp-Session-Id)
HTTP+SSE	Full. Deprecated by the specification; prefer Streamable HTTP

Paginated tools/list, resources/list, resources/templates/list, and prompts/list results are followed to the last page. A snapshot stopped early by the page bound records x-mcp-parser-incomplete rather than presenting a partial list as complete.

Not implemented: tools/call, resources/read, prompts/get, completions, and the client features the specification deprecated in 2026-07-28 (roots, sampling, logging). This package describes a server's surface; it does not exercise it.

Install

npm install mcp-parser

Quick Start

import { snapshot, validate, generateMarkdown } from "mcp-parser";
import { writeFile } from "node:fs/promises";

const spec = await snapshot({
  transport: { type: "stdio", command: "node", args: ["server.js"] },
});

const result = validate(spec);
if (!result.valid) {
  for (const d of result.diagnostics) {
    console.error(`${d.severity}: ${d.path} - ${d.message}`);
  }
}

await writeFile("mcp.json", JSON.stringify(spec, null, 2));
await writeFile("mcp.md", generateMarkdown(spec));

What is mcp.json?

A static snapshot of an MCP server's capabilities: its tools, resources, and prompts. Think of it as openapi.json for MCP servers.

MCP servers describe themselves at runtime via tools/list, resources/list, and prompts/list. An mcp.json captures that live surface in a versionable file for documentation, validation, diffing, and tooling that should not need a running server.

See mcp-schema for the full type definitions and JSON Schema.

API

parse(path, options?)

Parse an mcp.json file into a typed McpSpec object.

const spec = await parse("./mcp.json");
console.log(spec.server.name);   // "my-server"
console.log(spec.tools?.length); // 5

Options:

• dereference (default: true). Resolves $ref pointers using the spec's $defs.

parseString(content, options?)

Parse a JSON string directly.

const spec = parseString('{ "mcpSpec": "0.3.1", ... }');

validate(spec)

Validate an McpSpec for correctness and best practices.

const result = validate(spec);
// result.valid: boolean (true if no errors)
// result.diagnostics: array of { severity, path, message }

Checks for:

• Required fields (mcpSpec, server, tool names, inputSchema)
• Duplicate tool/resource/prompt names
• Missing descriptions (warnings)
• Invalid inputSchema types

snapshot(options)

Connect to a running MCP server and capture a static snapshot over any of the three MCP transports.

import { snapshot } from "mcp-parser";
import { writeFile } from "node:fs/promises";

// stdio
const spec = await snapshot({
  transport: { type: "stdio", command: "node", args: ["server.js"] },
});

// SSE
const spec = await snapshot({
  transport: { type: "sse", url: "http://localhost:3000/sse" },
});

// Streamable HTTP
const spec = await snapshot({
  transport: { type: "streamable-http", url: "http://localhost:3000/mcp" },
});

await writeFile("mcp.json", JSON.stringify(spec, null, 2));

All transports support an optional timeout (default: 30s). SSE and HTTP transports accept a headers object for authentication.

generateMarkdown(spec)

Generate a full markdown reference document.

generateLlmsTxt(spec, baseUrl?)

Generate a compact llms.txt-style index. This is a compatibility export for tools and docs sites that already consume the convention, not a guarantee that model providers will discover or fetch it automatically.

const txt = generateLlmsTxt(spec, "https://docs.example.com");

generateLlmsFullTxt(spec)

Generate a complete markdown reference with the server context inline.

const full = generateLlmsFullTxt(spec);

CLI

# Parse and pretty-print
mcp-parser parse ./mcp.json

# Validate
mcp-parser validate ./mcp.json

# Snapshot via stdio
mcp-parser snapshot --stdio "node server.js" -o mcp.json

# Snapshot via SSE
mcp-parser snapshot --sse http://localhost:3000/sse -o mcp.json

# Snapshot via streamable HTTP
mcp-parser snapshot --http http://localhost:3000/mcp -o mcp.json

# With auth headers
mcp-parser snapshot --sse http://localhost:3000/sse --header "Authorization:Bearer tok" -o mcp.json

# Generate markdown reference (default)
mcp-parser generate ./mcp.json -o mcp.md

# Generate compact context index
mcp-parser generate ./mcp.json --format llms-txt -o llms.txt

# Generate full context reference
mcp-parser generate ./mcp.json --format llms-full-txt -o llms-full.txt

MCP Specification Resources

• MCP Specification (current revision)
• MCP Specification 2025-11-25 (the revision this client requests)
• Specification repo (includes JSON Schema for each protocol version)
• TypeScript SDK (@modelcontextprotocol/sdk)
• Python SDK (mcp on PyPI)

Related

• mcp-schema: TypeScript types and JSON Schema for MCP specs
• sourcey: generate documentation from MCP specs, OpenAPI, and markdown

License

MIT