profilarr-mcp AGENTS.md

AGENTS.md — profilarr-mcp

MCP server exposing Profilarr's v1 REST API (/api/v1) as tools so an LLM can inspect and manage a Profilarr instance: linked databases, connected Radarr/Sonarr instances, backups, jobs, announcements, and system status. Uses FastMCP, uv for deps.

Exposed as 4 resource-scoped portmanteau tools, not one tool per endpoint — see "Portmanteau registration" below. A prior version registered all 20 endpoints individually.

Testing

• Offline suite: make test (or uv run pytest)
• Live integration (needs PROFILARR_URL/PROFILARR_API_KEY): make test-integration

Release workflow

Always use the make bump-* targets to bump the version (uv version --bump patch|minor|major), which updates pyproject.toml and uv.lock together. Do NOT edit the version by hand.

• Bump: make bump-patch (or bump-minor / bump-major)
• Commit message is just the version, e.g. 0.1.2 — nothing else.
• Tag it v<version> (e.g. v0.1.2).
• Push main and the tag:

  git push origin main
  git push origin v<version>
  

• Then sync the project copy:

  cd /home/savagecore/Documents/christopfarr/mcp/profilarr-mcp
  git fetch origin && git reset --hard origin/main
  

• Deploy to the Proxmox host (root SSH key): pull the repo then reinstall the uv tool:

  ssh [email protected] -- 'cd /root/profilarr-mcp && git fetch origin && git reset --hard origin/main'
  ssh [email protected] -- 'cd /root/profilarr-mcp && uv tool install --force .'
  

  The host runs it via uv tool install → /root/.local/bin/profilarr-mcp (not from the repo).

Env note

Profilarr's v1 programmatic API is deliberately small. All tools are keyed on PROFILARR_API_KEY; keep API keys out of code, logs, and commit messages.

Portmanteau registration — do not go back to one tool per endpoint

• _GROUPS near the bottom of profilarr_mcp.py buckets every endpoint function into one of 4 resource groups (profilarr_databases, profilarr_backups, profilarr_jobs_arr, profilarr_meta). _register_tools() registers exactly one MCP tool per group via _register_group, which wraps the group's functions in a single dispatch(operation, arguments) closure. The endpoint functions themselves are unchanged — they're plain callables looked up by name via globals(), not separately-registered tools.
• operation is typed Literal[<the group's function names>], so FastMCP/pydantic validates it against the real operation list before dispatch ever runs — an invalid operation never reaches the group tool's body.
• Adding a new endpoint: write the function as before (no decorator), then add its name to exactly one group in _GROUPS. tests/test_tools.py::test_all_operations_grouped fails if a name doesn't resolve to a real module attribute.
• New resource area big enough to need its own group: add a new _GROUPS key. Keep the total group count at or under ~15.
• If you're tempted to add a per-endpoint @mcp.tool decorator back, don't — every endpoint must be reachable only via its group's operation enum.
• Annotations: a group tool is readOnlyHint=True (READONLY — a module constant added during the portmanteau refactor since the file previously only had a bare DESTRUCTIVE constant and used inline ToolAnnotations(readOnlyHint=True) per read-only tool) only when every operation in it was originally read-only. Mixed groups carry no hints.