feat(mcp_registry): #201 PR 3 of 5. Filesystem install/uninstall + LockBackend lease + serializer + CLI

CLAUDE.md

@@ -207,7 +207,7 @@ uv run pytest                            # full suite
 uv run pytest tests/test_<module>.py -v  # one module
 ```
 
-Run `uv run pytest --collect-only -q | tail -1` for the live test count (last refresh: 3,199 tests collected, 2026-06-03). New backend protocols add ~25 conformance + ~10 impl-specific tests. New features ship with tests. Migration-shaped PRs need parameterized fixture tests across the backend protocol — the conformance suite is what keeps the protocol honest.
+Run `uv run pytest --collect-only -q | tail -1` for the live test count (last refresh: 3,232 tests collected, 2026-06-04). New backend protocols add ~25 conformance + ~10 impl-specific tests. New features ship with tests. Migration-shaped PRs need parameterized fixture tests across the backend protocol — the conformance suite is what keeps the protocol honest.
 
 ### Releases + SemVer
 
@@ -292,7 +292,7 @@ If the project ever needs to optimize differently, `docs/methodology.md` is the
 | Doc | Purpose |
 |-----|---------|
 | `docs/architecture.md` | Mental model in diagrams. Read first. |
-| `docs/spec/01-...33-persona-backend.md` | Locked spec (33 docs today, 31 locked + 2 drafts at spec/26 (cascade bundle) and spec/30 (responsibility audit)). The product. |
+| `docs/spec/01-...36-mcp-server-registry-backend.md` | Locked spec (35 docs today, 31 locked + 4 drafts at spec/26 (cascade bundle), spec/30 (responsibility audit), spec/35 (init wizard), and spec/36 (MCPServerRegistryBackend)). The product. |
 | `docs/implementation/` | Build guides per runtime (cron, Claude skill, dashboard) |
 | `docs/deployment/versioning.md`, `upgrading.md` | SemVer + operator runbook |
 | `docs/deployment/release-runbook.md` | Maintainer `/ship` runbook: two-mode workflow + manual surface check |
@@ -341,7 +341,7 @@ These are not forbidden forever — they're explicitly deferred with rationale.
 
 ## Status
 
-**v0.13.0, alpha, PUBLIC.** Core runtime stable. Test suite: run `uv run pytest --collect-only -q | tail -1` for the live count (last refresh: 3,199 tests collected, 2026-06-03). Capability-gated skips fall into four buckets — ToolRegistry conformance (filesystem-shape + `supports_uninstall=False` variants), AgentProfile (skill-content + filesystem-shape on SQLite), cross-process Redis (require real Redis instead of fakeredis), and judge-conformance dispatch (LLM-only + PolicyJudge concurrent-evaluate). Full CI runs against `uv sync --extra dev --extra openai --extra validation --extra redis`. **Eleven backend protocols shipped**:
+**v0.13.0, alpha, PUBLIC.** Core runtime stable. Test suite: run `uv run pytest --collect-only -q | tail -1` for the live count (last refresh: 3,232 tests collected, 2026-06-04). Capability-gated skips fall into four buckets — ToolRegistry conformance (filesystem-shape + `supports_uninstall=False` variants), AgentProfile (skill-content + filesystem-shape on SQLite), cross-process Redis (require real Redis instead of fakeredis), and judge-conformance dispatch (LLM-only + PolicyJudge concurrent-evaluate). Full CI runs against `uv sync --extra dev --extra openai --extra validation --extra redis`. **Eleven backend protocols shipped**:
 
 - **MemoryBackend** (PR #57) — filesystem reference impl + conformance suite.
 - **LLMBackend** (#87) — Anthropic + OpenAI + Moonshot reference impls, registered at framework import; conformance suite parametrizes across all three.

README.md

@@ -182,7 +182,7 @@ Start at [`docs/README.md`](docs/README.md) for the spec entry point. The locked
 - [33 — PersonaBackend Protocol](docs/spec/33-persona-backend.md) — persona ownership, snapshot/restore, `persona.link.md` format
 - [34 — CorpusBackend Protocol](docs/spec/34-corpus-backend.md) — wiki/raw corpus protocol; filesystem + SQLite (FTS5) reference impls; GB-scale indexed full-text search
 - [35 — init wizard](docs/spec/35-init-wizard.md) — `atomic-agents init` on-ramp; template scaffolding + Add-to-it merge; CI-friendly `--from-template` (RFC)
-- [36 — MCPServerRegistryBackend Protocol](docs/spec/36-mcp-server-registry-backend.md) — MCP server catalog + install/audit; `FilesystemMCPServerRegistryBackend` reference impl; `atomic-agents mcp-registry` CLI (DRAFT, PR 1 of 5)
+- [36 — MCPServerRegistryBackend Protocol](docs/spec/36-mcp-server-registry-backend.md) — MCP server catalog + install/audit; `FilesystemMCPServerRegistryBackend` reference impl; `atomic-agents mcp-registry` CLI (DRAFT, PR 3 of 5)
 
 Each spec doc is locked when the implementation matches and tests pass. Spec changes that imply implementation changes get filed as GitHub issues. **Spec docs separate shipped behavior from explicit future / deferred boundaries** — sections that describe behavior not yet implemented are explicitly marked as such, not silently aspirational.
 
@@ -282,7 +282,7 @@ Same pattern for OpenAI (`atomic-agents-openai`) and Moonshot (`atomic-agents-mo
 ## Repository structure
 
 - `atomic_agents/` — the Python package (runtime in `agent.py`; backend protocols in `memory/`, `_llm.py`, `_locks.py`, `_costs.py`, etc.; CLI in `cli.py`; preflight in `doctor.py`)
-- `tests/` 3199 tests collected (3141 passing + 58 skipped), Python 3.11 + 3.12 matrix
+- `tests/` 3,232 tests collected, Python 3.11 + 3.12 matrix
 - `docs/` — [spec entry point](docs/README.md), [`architecture.md`](docs/architecture.md), [`spec/`](docs/spec/) (31 locked docs + 4 RFCs/DRAFTs), [`deployment/`](docs/deployment/) (8 operator runbooks), [`samples/caldwell/`](docs/samples/caldwell/) (complete worked example), [`GOVERNANCE.md`](docs/GOVERNANCE.md), [`TENSIONS.md`](docs/TENSIONS.md), [`methodology.md`](docs/methodology.md)
 - `extras/` — operational templates (Claude Code skill wrappers, macOS LaunchAgent plists, cron examples)
 
@@ -313,4 +313,4 @@ Before opening a PR, read [`CLAUDE.md`](CLAUDE.md) (the project's design ethos a
 
 ## Status
 
-**v0.13.0, alpha.** Core runtime stable. 3199 tests collected (3141 passing + 58 skipped) on Python 3.11 / 3.12. Eleven of twelve backend protocols shipped (see the backend protocols table above); `MCPServerRegistryBackend` planned. The surface stabilizes at v1.0. Pre-1.0 — Minor releases may contain breaking changes (see [`docs/deployment/versioning.md`](docs/deployment/versioning.md)). Single-maintainer project; reference implementation anyone can use, fork, or extend.
+**v0.13.0, alpha.** Core runtime stable. 3,232 tests collected on Python 3.11 / 3.12. Eleven of twelve backend protocols shipped (see the backend protocols table above); `MCPServerRegistryBackend` in progress (PR 3 of 5). The surface stabilizes at v1.0. Pre-1.0 — Minor releases may contain breaking changes (see [`docs/deployment/versioning.md`](docs/deployment/versioning.md)). Single-maintainer project; reference implementation anyone can use, fork, or extend.

atomic_agents/_io.py

@@ -124,6 +124,11 @@ def cleanup_stale_tempfiles(directory: Path) -> list[Path]:
     """Find and delete leftover .tmp files from crashed writes.
 
     Run as part of lint or startup. Returns the list of files cleaned.
+
+    Note: this uses rglob which scans the entire directory tree recursively.
+    Prefer cleanup_stale_tempfiles_for_file when the target file is known;
+    that function is scoped to siblings of a specific file rather than the
+    full tree.
     """
     cleaned = []
     if not directory.exists():
@@ -135,3 +140,28 @@ def cleanup_stale_tempfiles(directory: Path) -> list[Path]:
         except FileNotFoundError:
             pass
     return cleaned
+
+
+def cleanup_stale_tempfiles_for_file(target: Path) -> list[Path]:
+    """Delete leftover .tmp files from crashed atomic_write calls for target.
+
+    Scoped to the parent directory of target, matching only tempfiles that
+    atomic_write would have created for this specific file. The pattern is
+    `.<name>.<random>.tmp` (the same pattern atomic_write uses when creating
+    the temporary file next to the destination).
+
+    Returns the list of files removed. Ignores FileNotFoundError races
+    (another process may have already cleaned the same file).
+
+    Use inside a lock so cleanup is serialized with the install/uninstall
+    operation that follows.
+    """
+    cleaned = []
+    pattern = f".{target.name}.*.tmp"
+    for path in target.parent.glob(pattern):
+        try:
+            path.unlink()
+            cleaned.append(path)
+        except FileNotFoundError:
+            pass
+    return cleaned