Why
MCP is how coding agents reach tools. The closest feature rival, @entro314labs/ai-changelog-generator, already ships an MCP server and a desktop extension — that is currently its clearest advantage.
src/index.ts already exports generate, promote, and PromptSession as a clean programmatic API. The server is a thin adapter over it, not new functionality.
Scope
Expose as MCP tools:
| Tool |
Maps to |
Notes |
generate_release_notes |
generate() |
range, environment, language, format |
promote_release_notes |
promote() |
returns files without writing by default |
ask_release_notes |
PromptSession |
read-only questions over existing notes |
list_releases |
output-path discovery |
what already exists on disk |
Design constraints:
- Read-only by default. Writing to disk is opt-in per call. An agent should be able to explore without side effects.
- Zod schemas for every tool input —
zod is already a dependency.
- Surface token usage in the response so the caller can see spend.
- Ship as a
bin entry so it runs under npx with no install.
Acceptance criteria
Files
New src/bin/mcp.ts, src/index.ts, package.json (bin), new docs/mcp.md
Why
MCP is how coding agents reach tools. The closest feature rival,
@entro314labs/ai-changelog-generator, already ships an MCP server and a desktop extension — that is currently its clearest advantage.src/index.tsalready exportsgenerate,promote, andPromptSessionas a clean programmatic API. The server is a thin adapter over it, not new functionality.Scope
Expose as MCP tools:
generate_release_notesgenerate()promote_release_notespromote()ask_release_notesPromptSessionlist_releasesoutput-pathdiscoveryDesign constraints:
zodis already a dependency.binentry so it runs undernpxwith no install.Acceptance criteria
docs/mcp.mdwith a copy-pasteable config blockFiles
New
src/bin/mcp.ts,src/index.ts,package.json(bin), newdocs/mcp.md