YouTube Research MCP v2.0.0

Publication status

Published to npm and the official MCP Registry. The MCPB bundle and SHA256SUMS are available below. Clean-room installation of the published npm package passed, including real public-video research and error handling. Optional Smithery publication was skipped because its API key is not configured.

Compatibility

Requires Node.js 22 or newer (previously Node.js 20). Upgrade Node before installing v2.0.0. Tool names and existing transcript workflows remain available; new pagination and provenance fields are additive.

Changes

• Keep the HTTP service alive after individual MCP session termination.
• Add source-native search pagination, comment filters and bounded reply retrieval with explicit coverage metadata.
• Preserve manual, automatic and translated caption provenance.
• Deduplicate replies and bound requests, reply counts and response size.
• Update dependencies and remove the vulnerable MCPB CLI signing dependency while retaining standard unsigned ZIP bundles and official schema validation.
• Improve setup documentation and repair the capability badge.

Validated with 44 tests, Node.js 22/24 CI, CodeQL, MCP Inspector, real public-video research and clean-room MCPB execution. Dependency audit: zero known vulnerabilities at validation time. Data API comment/search pagination was tested with provider stubs.

Includes all contributions integrated in #49.

YouTube Research MCP v1.2.0

YouTube Research MCP v1.2.0 makes citation-ready evidence self-identifying and completes one-click desktop distribution.

Highlights

• Every research-video and research-videos result now includes video title, channel identity, canonical URLs, thumbnail, and readable citation labels such as Introducing GPT-5 [00:35].
• Transcript segments and source identity are obtained from one YouTube information request and cached together.
• A reproducible MCPB pipeline vendors production dependencies, unpacks the artifact in a clean directory, launches the bundled MCP server, and verifies all 14 tools.
• GitHub Releases and Smithery can distribute the local one-click bundle for Claude Desktop and other MCPB-compatible clients.

Install

codex mcp add youtube-research -- npx -y @coyasong/youtube-mcp-server@latest

No YouTube API key is required for transcript research.

Verification

• Node 20/22 CI
• 14 automated tests
• live citation test against Introducing GPT-5
• published-package virtual-user smoke test
• clean-room MCPB runtime smoke test
• Docker production build
• npm audit: 0 vulnerabilities

YouTube Research MCP v1.1.1

Patch release correcting the case-sensitive publisher namespace used by the official MCP Registry.

One-click desktop install

Download youtube-research-mcp.mcpb and open it in Claude Desktop or another MCPB-compatible client. The 8.5 MB bundle includes its production dependencies and runs locally without a YouTube API key for transcript research.

The release asset is built reproducibly, unpacked in a clean directory, launched as a real MCP server, and checked for all 14 tools before upload.

Fixed

• Changed the ownership marker from io.github.coyasong/youtube-research to the authorized GitHub namespace io.github.coyaSONG/youtube-research.
• Synchronized npm, runtime, changelog, and Registry metadata at version 1.1.1.

Install with npm

codex mcp add youtube-research -- npx -y @coyasong/youtube-mcp-server@latest

• npm: https://www.npmjs.com/package/@coyasong/youtube-mcp-server/v/1.1.1
• Official MCP Registry name: io.github.coyaSONG/youtube-research

Node 20/22 CI, 12 automated tests, MCPB clean-room verification, the published-package virtual-user smoke test, npm audit, and Docker build pass.

YouTube Research MCP v1.1.0

YouTube MCP Server is now YouTube Research MCP: a citation-first research tool for Codex, Claude, Cursor, VS Code, and any MCP client.

Highlights

• Research transcripts without a YouTube API key.
• Open every citation at the exact timestamp with research-video.
• Compare focused evidence across 2–5 videos with research-videos.
• Return 98.9% fewer response characters in the included focused-query benchmark (970 vs. 85,518).
• Run locally over stdio with npx, or deploy Streamable HTTP with optional bearer authentication.
• Use transcript-only or full YouTube Data API modes.
• Install with Codex, VS Code, Smithery, Docker, or any stdio/HTTP MCP client.

Install

codex mcp add youtube-research -- npx -y @coyasong/youtube-mcp-server@1.1.0

Package: https://www.npmjs.com/package/@coyasong/youtube-mcp-server/v/1.1.0

Reliability and security

• Replaced the unreliable caption scraper with YouTube.js caption-track retrieval.
• Added shared bounded transcript caching, real analysis segmentation, structured MCP output, and read-only tool annotations.
• Added Node 20/22 CI, 11 automated tests, a two-video live smoke test, Dependabot, security policy, and non-root Docker runtime.
• Added optional constant-time bearer authentication, CORS restriction, session capacity limits, and idle cleanup for remote deployments.

Verification

• npm test: 11/11 passing
• npm run test:live: passing against two public videos
• npm audit --audit-level=high: 0 vulnerabilities
• publint: all good
• Published-package MCP handshake: 14 tools discovered

See CHANGELOG.md for the full change list.