docs: add CONTRIBUTORS.md issues-only policy (#1517)

Document the contribution policy: we accept issues, not pull requests.
Maintainers handle design and implementation through a prompt-driven
workflow. Contributors who have already built a change locally should
share the prompt they used rather than a diff.

Cross-reference the policy from AGENTS.md.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
cliffhall 2026-06-27 18:47:51 -04:00
commit 73d7439320
2 changed files with 86 additions and 0 deletions

View file

@ -73,6 +73,13 @@ After installing, `npm run build` builds all clients. The launcher scripts (`npm
- v1.5 - https://github.com/orgs/modelcontextprotocol/projects/39
- v1 - https://github.com/orgs/modelcontextprotocol/projects/11
## Contributing
External contributions are accepted as **issues, not pull requests** —
maintainers handle design and implementation through a prompt-driven workflow.
If you've already built a change locally, share the **prompt** you used, not a
diff. See [`CONTRIBUTORS.md`](./CONTRIBUTORS.md) for the full policy.
## Project Status and Direction
* The main branch currently contains the legacy version of the Inspector, which we are accepting bug fixes and minor improvement PRs for.

79
CONTRIBUTORS.md Normal file
View file

@ -0,0 +1,79 @@
# Contributing to MCP Inspector
Thank you for your interest in improving the MCP Inspector. Contributors are
genuinely valued — the goal of this document is to channel your input into a
form we can act on quickly and consistently.
## TL;DR
**We accept issues, not pull requests.** Design and implementation are done by
the maintainers. If you've already built a fix or feature locally, share **the
prompt you used** to produce it — not the source code.
## Why this policy exists
The Inspector v2 is developed with an AI-assisted, prompt-driven workflow built
around a consistent architecture, a shared design system, and strict testing
gates (see [`AGENTS.md`](./AGENTS.md)). Every component follows the same
conventions: "dumb" components that take data and callbacks as props, Mantine
for all UI, theme variants instead of ad-hoc CSS, and a uniform per-file
coverage gate of ≥ 90% on lines, statements, functions, and branches.
Accepting raw source PRs creates friction: a diff written outside this pipeline
has to be reverse-engineered to fit our component/theme conventions, coverage
gates, and review process — often it's faster to re-derive the change than to
adapt the patch. Capturing your **intent** (a well-formed issue) or the
**prompt** that generated your local change lets us reproduce the work inside
our own workflow and standards, with the quality bar already baked in.
This policy is about efficiency, not gatekeeping. Your ideas, bug reports, and
prompts directly shape what gets built.
## How to contribute a bug report or feature request
Open a well-scoped issue on the appropriate version board. The Inspector
maintains three lines of development, each with its own board and base branch:
| Target | Board | Base branch | Label |
| --- | --- | --- | --- |
| v2 (current focus) | [v2 board](https://github.com/orgs/modelcontextprotocol/projects/28) | `v2/main` | `v2` |
| v1.5 | [v1.5 board](https://github.com/orgs/modelcontextprotocol/projects/39) | `v1.5/main` | `v1.5` |
| v1 (legacy) | [v1 board](https://github.com/orgs/modelcontextprotocol/projects/11) | `main` | `v1` |
Most new work targets **v2** — open it on the v2 board and label it `v2`. Apply
the label that matches the target version (`v1` / `v1.5` / `v2`) so the issue
can be filtered correctly; see the "Label by version" convention in
[`AGENTS.md`](./AGENTS.md).
## If you've already fixed it locally
Please don't send a diff or open a pull request. Instead, open an issue that
includes:
- **The prompt(s) you used** to generate the change — the exact text, so we can
reproduce it through our own workflow.
- **A description of the behavior before and after** your change.
- **How you verified it** (steps you ran, tests you added, what you observed).
We'll reproduce the change through our pipeline so it lands with the right
conventions, tests, and coverage.
## What makes a good issue or prompt submission
A great submission gives us everything we need to act without a round-trip:
- **Clear reproduction or use case** — exact steps to reproduce a bug, or a
concrete description of the feature and the problem it solves.
- **Expected vs. actual behavior** — what you saw, and what you expected
instead.
- **Affected client** — which incarnation is involved: **Web**, **TUI**, or
**CLI** (or "all" / "core" if it's shared logic).
- **Environment details** when relevant — OS, Node version, the MCP server you
were inspecting, and any relevant config.
- **The exact prompt text**, if you generated a local change and want us to
reproduce it.
## Questions
If you're unsure which board or label applies, open the issue on the v2 board
and say so — we'll route it. Thanks for helping make the Inspector better.