Project-local AI tooling config (MCP servers + equivalents) checked into the repo

## Why
Modern AI coding tools support project-scoped config. Claude Code uses MCP servers in `.mcp.json` at the repo root; Cursor has project rules, Aider has `.aider.conf.yml`, etc. Some configs are genuinely project-wide (a nuget-aware MCP for dependency queries, a milestone-aware MCP for issue triage, a secrets-store MCP once ADR-0003 lands) rather than per-machine personal config.

If these stay in personal config, every new maintainer rediscovers and replicates the setup. Checking the project-useful ones into the repo gives every maintainer on the same tool the same baseline automatically.

## Action
- [ ] Identify tools/servers that benefit every maintainer, not just the person who set them up:
- Anything reading project-local state (issues, milestones, ADRs, the architecture brief)
- Anything interacting read-only with project external systems (NuGet feed, GitHub Actions)
- Anything enforcing conventions (e.g. a CODEOWNERS-aware reviewer-suggestion server)
- [ ] Check appropriate config into the repo:
- `.mcp.json` for Claude Code
- Equivalents for Cursor, Aider, others as the maintainer mix evolves
- [ ] Document in AGENTS.md — "if you use , the project provides these pre-configured servers/extensions"
- [ ] Discipline: "shared config must benefit the project, not be personal preference" — each checked-in config needs a one-line justification
- [ ] Decide whether any config needs secrets (API keys) and document the secret-provisioning pattern

## Coordinate with
- AGENTS.md adoption issue — referenced from there
- Bot identity issue — some MCP servers might run as the project's bot identity rather than each maintainer's account

## Done when
- At least one project-useful AI tool config is checked into the repo and documented
- AGENTS.md tells contributors how to pick up the pre-configured environment for their tool
- A discipline exists for adding new configs (justification required)