MCP Server

Use Releases as an AI agent tool server via the Model Context Protocol.

Pair with agent skills

One-line setup so your agent reaches for these tools on its own. See the skills page for the full list.

npx skills add buildinternet/releases-cli

Drops skill files into the project. Works in Claude Code, Codex, Cursor, OpenCode.

Remote server (recommended)

Connect to the hosted MCP server at https://mcp.releases.sh/mcp. No installation or API keys required — all tools are read-only and public.

Setup instructions

General

The hosted MCP server supports Streamable HTTP at:

https://mcp.releases.sh/mcp

Use that URL directly in clients with native remote MCP support. For clients that only support stdio MCP servers, use mcp-remote as a compatibility bridge.

One-click install

Click to install in a supported editor. The deeplink opens the app and prompts you to confirm before adding the server.

Cursor VS Code

Claude Code

claude mcp add --transport http releases https://mcp.releases.sh/mcp

Codex

codex mcp add releases --url https://mcp.releases.sh/mcp

VS Code, Windsurf, Zed, and others

For clients without native remote MCP support, use mcp-remote:

{
  "mcpServers": {
    "releases": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://mcp.releases.sh/mcp"]
    }
  }
}

Available tools

Read tools

Read-only tools available on the remote server with no authentication.

Identifiers: every tool that takes an org / product / source identifier accepts the typed ID (org_…, prod_…, src_…) interchangeably with the slug. Source and product params also accept an org/slug coordinate (e.g. vercel/nextjs). Releases are addressed by id only — get_release takes a rel_… id or a bare 21-char nanoid, with no slug form. Each tool's inputSchema.description lists the concrete shapes it accepts.

Tool	Description
`search`	Unified search across orgs, catalog (products + standalone sources), and release content. Accepts `type: ("orgs" \| "catalog" \| "releases")[]` to skip sections; `mode: "lexical" \| "semantic" \| "hybrid"` (default `hybrid`) for release retrieval; `entity` to scope releases; `since` / `until` to bound release hits by publish date (ISO or `90d`/`4w`/`6m`/`2y`). Release hits carry a `kind: "release" \| "changelog_chunk"` discriminator; chunk hits include `chunkOffset` and `chunkLength` so you can chain into `get_catalog_entry` with changelog slicing params for surrounding context.
`list_catalog`	Products and standalone sources folded into one list, each row tagged with an `entryType: "product" \| "source"` discriminator.
`get_catalog_entry`	Detail for a single catalog entry — product or source. Accepts slug or `prod_` / `src_` id. Source entries list tracked CHANGELOG files (path + byte size) by default. Pass `include_changelog: true` to inline the root CHANGELOG, or `changelog_path` / `changelog_offset` / `changelog_limit` / `changelog_tokens` to target a specific file or slice. Heading-aligned slicing supports monorepo per-package files; token-mode responses include `totalTokens` and `sliceTokens` for LLM budgeting (brackets: 2000/5000/10000/20000).
`get_latest_releases`	Get the most recent releases, optionally filtered by product, organization, release `type`, or a `since` / `until` publish-date window (ISO or `90d`/`4w`/`6m`/`2y`).
`get_release`	Fetch the full content of a single release by id. Accepts a `rel_` prefix or a bare nanoid.
`list_organizations`	List all organizations, searchable by name, slug, domain, or account handle.
`get_organization`	Detailed view of a single organization including accounts, tags, sources, products, and domain aliases. Shows a preview of the AI-generated overview by default; pass `include_overview: true` to inline the full briefing (with a stale warning if it's older than 30 days).
`search_releases` (dep.)	Deprecated shim — prefer `search` with `type: ["releases"]`. Same hybrid-retrieval shape kept for one release cycle.
`search_registry` (dep.)	Deprecated shim — prefer `search` with `type: ["orgs", "catalog"]`.
`list_sources` (dep.)	Deprecated — prefer `list_catalog`.
`list_products` (dep.)	Deprecated — prefer `list_catalog`.
`get_product` (dep.)	Deprecated — prefer `get_catalog_entry`.

On-demand GitHub lookup

search and search_releases fall back to an on-demand GitHub lookup when the query is a {org}/{repo} coordinate and the in-index search returns no hits. The result is merged under a lookup field in the tool response so the agent can see the repo state without a second tool call.

Field	Description
`status`	`indexed` (newly materialized), `existing` (already tracked), `empty` (real repo, no releases or CHANGELOG yet), `not_found` (no public repo), or `deferred` (GitHub rate-limit or 5xx — try shortly).
`source`	Source record for the materialized or existing repo. Present on `indexed`, `existing`, and `empty`.
`releases`	Inline release preview. Present on `indexed` and `existing`.
`relatedOrg`	"Did you mean" rail — set when the org segment matches a known org but the specific repo doesn't. Lists the org and up to 5 sibling sources.

lookup is null when the query is not coordinate-shaped or when existing search hits were found. Materialized rows are hidden (discovery: "on_demand"); a second search for the same coordinate resolves through the normal cache path. Embeddings still run for on-demand sources, so semantic search picks them up on the second hit; AI features (overviews, summarization, playbook regen) skip them.

Example usage

Once configured, you can ask Claude to interact with the release index directly:

"What did Vercel ship last week?"
"Search for breaking changes in the Prisma changelog"
"Compare Next.js and Remix releases from the last 30 days"
"Summarize Cloudflare's recent releases, focusing on Workers"