uxarray-mcp 0.3.0__tar.gz → 2026.9.0__tar.gz

{uxarray_mcp-0.3.0 → uxarray_mcp-2026.9.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: uxarray-mcp
-Version: 0.3.0
+Version: 2026.9.0
 Summary: MCP server for analyzing unstructured meshes with UXarray
 Keywords: uxarray,mcp,unstructured grids,scientific computing,globus compute
 Author: Rajeev Jain, Dayan Abdulla
@@ -218,17 +218,18 @@ Classifier: Programming Language :: Python :: 3.13
 Classifier: License :: OSI Approved :: Apache Software License
 Requires-Dist: toolregistry-server[mcp]>=0.5.0
 Requires-Dist: toolregistry>=0.16.0
-Requires-Dist: mcp>=1.24,<3
+Requires-Dist: mcp>=1.27,<3
 Requires-Dist: holoviews>=1.19.0
 Requires-Dist: matplotlib>=3.9.0
-Requires-Dist: pyyaml>=6.0
-Requires-Dist: uxarray>=2026.8.0
+Requires-Dist: pyyaml>=6.0.1
+Requires-Dist: uxarray>=2026.9.0,<2026.10
 Requires-Dist: sphinx>=7.0 ; extra == 'docs'
 Requires-Dist: sphinx-book-theme>=1.1.0 ; extra == 'docs'
 Requires-Dist: myst-parser>=3.0 ; extra == 'docs'
 Requires-Dist: academy-py>=0.3.1 ; extra == 'hpc'
 Requires-Dist: globus-compute-sdk>=4.5.0 ; extra == 'hpc'
 Requires-Dist: toolregistry-server[openapi]>=0.5.0 ; extra == 'openapi'
+Requires-Dist: globus-cli>=3.30 ; extra == 'transfer'
 Requires-Python: >=3.11
 Project-URL: Documentation, https://uxarray-mcp-server.readthedocs.io
 Project-URL: Source, https://github.com/UXARRAY/uxarray-mcp-server
@@ -236,12 +237,13 @@ Project-URL: Tracker, https://github.com/UXARRAY/uxarray-mcp-server/issues
 Provides-Extra: docs
 Provides-Extra: hpc
 Provides-Extra: openapi
+Provides-Extra: transfer
 Description-Content-Type: text/markdown
 
 # UXarray MCP Server
 
-An MCP server that lets an AI assistant (Claude Code, Claude Desktop, Cursor,
-or any MCP client) analyze unstructured climate meshes with
+An MCP server that lets an AI assistant (Claude Code, Claude Desktop, Codex,
+opencode, Cursor, or any MCP client) analyze unstructured climate meshes with
 [UXarray](https://uxarray.readthedocs.io/) — locally on your machine, or
 remotely on an HPC system you have access to.
 
@@ -288,6 +290,15 @@ You are most likely one of:
 5. **Just trying it out, or running an agent harness** — you don't want to
    install a scientific Python stack at all. → [Docker](#docker).
 
+**Want to see what this is for before installing anything?** Read the worked
+case study: [**a 10-year CONUS precipitation map from 37 GB that never
+moved**](case-studies/conus-precipitation-gdex/README.md). One paragraph of
+plain English on a laptop, ten years of 6-hourly CESM output on the NCAR
+filesystem, all the compute on a Casper worker, and a 178 KB PNG plus an
+auditable provenance record coming back. It covers what MCP is, what actually
+ran where, why it took 7.7 minutes, what it cost in tokens, and how to set the
+same thing up yourself.
+
 ---
 
 ## Docker
@@ -374,16 +385,22 @@ uv tool install --python 3.12 uxarray-mcp
 
 # Or from a fresh clone (developer path)
 git clone https://github.com/UXARRAY/uxarray-mcp-server.git
-cd uxarray-mcp-server && uv sync --python 3.12
-# or: bash SETUP.sh   (does the sync + runs the local test suite in one step)
+cd uxarray-mcp-server && uv sync --python 3.12 --extra hpc --extra transfer
+# or: bash SETUP.sh   (local-only sync + runs the local test suite in one step)
 ```
 
-> **Why `--python 3.12`?** The server uses Globus Compute to submit work to
-> HPC endpoints, and Globus Compute's serializer is fragile across Python
-> minor versions — a 3.13 submitter against a 3.12 endpoint worker raises
-> `WorkerLost` on non-trivial payloads. HPC sites broadly ship 3.12 conda
-> stacks today, so we pin the install to match. Tracking removal of this pin
-> at [globus/globus-compute#2139](https://github.com/globus/globus-compute/issues/2139).
+The `hpc` and `transfer` extras hold the Globus Compute and Globus Transfer
+SDKs. Leave them off for a laptop-only install. Note that `uv sync` installs
+*exactly* the requested set: running a plain `uv sync` later removes the
+extras again, and `hpc/endpoint_status` will then report `unreachable` with
+`No module named 'globus_compute_sdk'`.
+
+> **Why `--python 3.12`?** Only the HPC path needs it. Globus Compute's
+> serializer is fragile across Python minor versions — a 3.13 submitter
+> against a 3.12 endpoint worker raises `WorkerLost` on non-trivial payloads,
+> and HPC sites broadly ship 3.12 conda stacks today. Local-only use works on
+> 3.11–3.13. Tracking removal of this constraint at
+> [globus/globus-compute#2139](https://github.com/globus/globus-compute/issues/2139).
 > `uv` downloads 3.12 automatically if your system doesn't have it.
 
 ### Step 2 — Write a starter config
@@ -395,16 +412,32 @@ uxarray-mcp setup
 Creates `~/.config/uxarray-mcp/config.yaml` with sensible defaults. Local mode
 needs nothing more.
 
+Three environment variables adjust where the server looks and writes:
+`UXARRAY_MCP_CONFIG` (path to a config file, checked before
+`~/.config/uxarray-mcp/config.yaml`), `UXARRAY_MCP_STATE_DIR` (sessions,
+result handles and rendered plots; default `~/.uxarray_mcp_server`), and
+`UXARRAY_MCP_VERDICT_POLICY` (`full`, `reference_only` or `off` for the
+postcondition block on every result).
+
 ### Step 3 — Connect your AI client
 
 **Claude Desktop**
 
 ```bash
-uxarray-mcp install-claude        # merges the mcpServers block into your config
+# merges the mcpServers block into the config file you name
+uxarray-mcp install-claude --config-path ~/Library/Application\ Support/Claude/claude_desktop_config.json
 # or
 uxarray-mcp install-claude --print-only   # prints the JSON to paste manually
 ```
 
+Without `--config-path` the command only prints the block; it never guesses
+where your Claude Desktop config lives.
+
+If you installed from a clone rather than `uv tool install`, the `uxarray-mcp`
+binary lives in the project `.venv`. In every client config below, use
+`uv --directory /path/to/uxarray-mcp-server run uxarray-mcp serve` as the
+command instead of a bare `uxarray-mcp serve`.
+
 Restart Claude Desktop. The `uxarray` server should appear in Settings →
 Developer.
 
@@ -416,10 +449,63 @@ claude mcp add uxarray --transport stdio -- uxarray-mcp serve
 
 Then `/mcp` in Claude Code; pick `uxarray`.
 
-**Cursor / other MCP clients**
+**Codex CLI**
+
+```bash
+codex mcp add uxarray -- uxarray-mcp serve
+```
+
+Or write `~/.codex/config.toml` directly:
+
+```toml
+[mcp_servers.uxarray]
+command = "uxarray-mcp"
+args = ["serve"]
+```
+
+Then `/mcp` in a Codex session.
 
-Add an MCP server entry pointing at `uxarray-mcp serve` over stdio. See your
-client's MCP docs.
+**opencode**
+
+Add to `~/.config/opencode/opencode.json`:
+
+```json
+{
+  "mcp": {
+    "uxarray": {
+      "type": "local",
+      "command": ["uxarray-mcp", "serve"],
+      "enabled": true
+    }
+  }
+}
+```
+
+The server registers 33 tools, which is a large tool schema to carry on every
+request. `"enabled": false` turns it off for sessions that are not doing mesh
+analysis.
+
+**Cursor**
+
+Add to `~/.cursor/mcp.json` (or a project-local `.cursor/mcp.json`):
+
+```json
+{
+  "mcpServers": {
+    "uxarray": {
+      "command": "uxarray-mcp",
+      "args": ["serve"]
+    }
+  }
+}
+```
+
+**Any other MCP client**
+
+The server speaks stdio, so every client wants the same two facts — the
+command `uxarray-mcp` and the argument `serve`. `uxarray-mcp install-claude
+--print-only` prints the `mcpServers` JSON block that most clients accept
+verbatim.
 
 ### Step 4 — Sanity check
 
@@ -476,16 +562,27 @@ Intent-shaped tools, not raw UXarray bindings — all local by default:
 - `get_capabilities` — what can I do with this mesh?
 - `analyze_dataset` — deterministic first-look: inspect, validate, area, zonal mean, plots.
 - `run_analysis` — one operation at a time (gradient, curl, subset, remap, …).
-- `plot_dataset` — mesh, geographic, variable, or zonal-mean plots.
+  Remaps take `method`: `nearest_neighbor`, `inverse_distance_weighted` or
+  `bilinear` on UXarray's own engine, or `conservative`, `nnn`, `dnn`,
+  `average` on [YAC](https://dkrz-sw.gitlab-pages.dkrz.de/yac/) (equivalently
+  `backend="yac"` with `yac_method`). Only `conservative` preserves the field
+  integral, and it needs YAC importable where the remap runs — build it with
+  `scripts/build_yac_local.sh` on a laptop or `scripts/hpc_build_yac.py` on a
+  worker, and put its `site-packages` on `PYTHONPATH`.
+- `plot_dataset` — `plot_type` of `mesh`, `mesh_geo`, `variable`, or `zonal_mean`.
 - `run_workflow`, `resume_workflow`, `get_status`, `get_result`, `manage_session` —
   persisted sessions and multi-step workflows.
 
-Full schema: [docs/tools.md](docs/tools.md).
+Helper namespaces also appear in `tools/list`: `session/*`, `hpc/*`,
+`io/list_datasets`, `contract/*` and `prompt/*`.
+
+Full schema and every `run_analysis` parameter: [docs/tools.md](docs/tools.md).
 
 **Protocol version.** We do not implement MCP directly; servers are built
 through `toolregistry-server`, which depends on the `mcp` Python SDK. As of
 `toolregistry-server` 0.5.0 and `toolregistry` 0.16.0 the SDK cap is lifted, so
-we resolve `mcp` 2.1.1 and negotiate spec **`2026-07-28`** (stateless core,
+we resolve `mcp` 1.27 or 2.x (2.1.1 at the last lock) and negotiate spec
+**`2026-07-28`** (stateless core,
 cacheable list results, MRTR). 0.16.0 also widens the recognized content-block
 set to audio, `resource_link`, and embedded resources.
 
@@ -508,10 +605,13 @@ auditable and the server actively flags common scientific pitfalls:
   the tool that ran, timestamp, input arguments, `execution_venue`
   (`local` or `hpc:<endpoint>`), and the UXarray/Python versions used.
 - **Derivative unit convention is never hidden.** `gradient`, `curl`, and
-  `divergence` echo `scale_by_radius` in both the result and provenance, so a
-  unit-sphere result can never be mistaken for a physical (per-metre) one.
-  Gradient and curl default to physical scaling, matching UXarray; pass
-  `scale_by_radius=False` explicitly for unit-sphere output.
+  `divergence` echo `scale_by_radius` and a `radius_basis` block (the radius
+  used and whether it came from the grid or the caller), so a unit-sphere
+  result can never be mistaken for a physical (per-metre) one. Most grid files
+  declare no `sphere_radius`; pass `sphere_radius=6371000` (metres) to attach
+  Earth's. Without it, or with `scale_by_radius=False`, the call is refused
+  with `outcome="input_required"` until you pass `acknowledge`. The same
+  `sphere_radius` argument turns `calculate_area` from steradians into m².
 - **Vector-calculus sanity guard.** `curl`/`divergence` warn (without blocking)
   when the two inputs are the same field, or when neither carries a
   velocity/flux-like `units` attribute — the classic "vorticity from two random
@@ -533,12 +633,14 @@ auditable and the server actively flags common scientific pitfalls:
 
 | Command | Purpose |
 |---|---|
-| `uxarray-mcp serve` | Run the MCP server (used by your AI client) |
+| `uxarray-mcp serve` | Run the MCP server (used by your AI client); `--profile core\|deferred-full`, `--transport stdio\|sse\|http` — see [docs/serving.md](docs/serving.md) |
+| `uxarray-mcp openapi` | Print the OpenAPI document for the HTTP transport |
 | `uxarray-mcp setup` | Write a starter config |
 | `uxarray-mcp endpoints add NAME UUID` | Register a Globus Compute endpoint |
-| `uxarray-mcp endpoints list` | Show configured endpoints |
+| `uxarray-mcp endpoints list` / `remove NAME` | Show or drop configured endpoints |
+| `uxarray-mcp transfer setup` | Check and fix everything a Globus Transfer needs, including the browser login an MCP server cannot open |
 | `uxarray-mcp doctor` | Validate local + (optionally) remote setup |
-| `uxarray-mcp install-claude` | Merge or print the Claude Desktop config block |
+| `uxarray-mcp install-claude --config-path FILE` | Merge (or `--print-only`) the Claude Desktop config block |
 
 ---
 
@@ -579,7 +681,7 @@ see **[SECURITY.md](SECURITY.md)**.
 ## Development
 
 ```bash
-uv sync --extra hpc --extra docs --dev
+uv sync --extra hpc --extra transfer --extra docs
 uv run pre-commit run --all-files
 uv run pytest tests/ --ignore=tests/test_remote_agent.py
 uv run sphinx-build -b html docs docs/_build/html

{uxarray_mcp-0.3.0 → uxarray_mcp-2026.9.0}/README.md

@@ -1,7 +1,7 @@
 # UXarray MCP Server
 
-An MCP server that lets an AI assistant (Claude Code, Claude Desktop, Cursor,
-or any MCP client) analyze unstructured climate meshes with
+An MCP server that lets an AI assistant (Claude Code, Claude Desktop, Codex,
+opencode, Cursor, or any MCP client) analyze unstructured climate meshes with
 [UXarray](https://uxarray.readthedocs.io/) — locally on your machine, or
 remotely on an HPC system you have access to.
 
@@ -48,6 +48,15 @@ You are most likely one of:
 5. **Just trying it out, or running an agent harness** — you don't want to
    install a scientific Python stack at all. → [Docker](#docker).
 
+**Want to see what this is for before installing anything?** Read the worked
+case study: [**a 10-year CONUS precipitation map from 37 GB that never
+moved**](case-studies/conus-precipitation-gdex/README.md). One paragraph of
+plain English on a laptop, ten years of 6-hourly CESM output on the NCAR
+filesystem, all the compute on a Casper worker, and a 178 KB PNG plus an
+auditable provenance record coming back. It covers what MCP is, what actually
+ran where, why it took 7.7 minutes, what it cost in tokens, and how to set the
+same thing up yourself.
+
 ---
 
 ## Docker
@@ -134,16 +143,22 @@ uv tool install --python 3.12 uxarray-mcp
 
 # Or from a fresh clone (developer path)
 git clone https://github.com/UXARRAY/uxarray-mcp-server.git
-cd uxarray-mcp-server && uv sync --python 3.12
-# or: bash SETUP.sh   (does the sync + runs the local test suite in one step)
+cd uxarray-mcp-server && uv sync --python 3.12 --extra hpc --extra transfer
+# or: bash SETUP.sh   (local-only sync + runs the local test suite in one step)
 ```
 
-> **Why `--python 3.12`?** The server uses Globus Compute to submit work to
-> HPC endpoints, and Globus Compute's serializer is fragile across Python
-> minor versions — a 3.13 submitter against a 3.12 endpoint worker raises
-> `WorkerLost` on non-trivial payloads. HPC sites broadly ship 3.12 conda
-> stacks today, so we pin the install to match. Tracking removal of this pin
-> at [globus/globus-compute#2139](https://github.com/globus/globus-compute/issues/2139).
+The `hpc` and `transfer` extras hold the Globus Compute and Globus Transfer
+SDKs. Leave them off for a laptop-only install. Note that `uv sync` installs
+*exactly* the requested set: running a plain `uv sync` later removes the
+extras again, and `hpc/endpoint_status` will then report `unreachable` with
+`No module named 'globus_compute_sdk'`.
+
+> **Why `--python 3.12`?** Only the HPC path needs it. Globus Compute's
+> serializer is fragile across Python minor versions — a 3.13 submitter
+> against a 3.12 endpoint worker raises `WorkerLost` on non-trivial payloads,
+> and HPC sites broadly ship 3.12 conda stacks today. Local-only use works on
+> 3.11–3.13. Tracking removal of this constraint at
+> [globus/globus-compute#2139](https://github.com/globus/globus-compute/issues/2139).
 > `uv` downloads 3.12 automatically if your system doesn't have it.
 
 ### Step 2 — Write a starter config
@@ -155,16 +170,32 @@ uxarray-mcp setup
 Creates `~/.config/uxarray-mcp/config.yaml` with sensible defaults. Local mode
 needs nothing more.
 
+Three environment variables adjust where the server looks and writes:
+`UXARRAY_MCP_CONFIG` (path to a config file, checked before
+`~/.config/uxarray-mcp/config.yaml`), `UXARRAY_MCP_STATE_DIR` (sessions,
+result handles and rendered plots; default `~/.uxarray_mcp_server`), and
+`UXARRAY_MCP_VERDICT_POLICY` (`full`, `reference_only` or `off` for the
+postcondition block on every result).
+
 ### Step 3 — Connect your AI client
 
 **Claude Desktop**
 
 ```bash
-uxarray-mcp install-claude        # merges the mcpServers block into your config
+# merges the mcpServers block into the config file you name
+uxarray-mcp install-claude --config-path ~/Library/Application\ Support/Claude/claude_desktop_config.json
 # or
 uxarray-mcp install-claude --print-only   # prints the JSON to paste manually
 ```
 
+Without `--config-path` the command only prints the block; it never guesses
+where your Claude Desktop config lives.
+
+If you installed from a clone rather than `uv tool install`, the `uxarray-mcp`
+binary lives in the project `.venv`. In every client config below, use
+`uv --directory /path/to/uxarray-mcp-server run uxarray-mcp serve` as the
+command instead of a bare `uxarray-mcp serve`.
+
 Restart Claude Desktop. The `uxarray` server should appear in Settings →
 Developer.
 
@@ -176,10 +207,63 @@ claude mcp add uxarray --transport stdio -- uxarray-mcp serve
 
 Then `/mcp` in Claude Code; pick `uxarray`.
 
-**Cursor / other MCP clients**
+**Codex CLI**
+
+```bash
+codex mcp add uxarray -- uxarray-mcp serve
+```
+
+Or write `~/.codex/config.toml` directly:
+
+```toml
+[mcp_servers.uxarray]
+command = "uxarray-mcp"
+args = ["serve"]
+```
+
+Then `/mcp` in a Codex session.
 
-Add an MCP server entry pointing at `uxarray-mcp serve` over stdio. See your
-client's MCP docs.
+**opencode**
+
+Add to `~/.config/opencode/opencode.json`:
+
+```json
+{
+  "mcp": {
+    "uxarray": {
+      "type": "local",
+      "command": ["uxarray-mcp", "serve"],
+      "enabled": true
+    }
+  }
+}
+```
+
+The server registers 33 tools, which is a large tool schema to carry on every
+request. `"enabled": false` turns it off for sessions that are not doing mesh
+analysis.
+
+**Cursor**
+
+Add to `~/.cursor/mcp.json` (or a project-local `.cursor/mcp.json`):
+
+```json
+{
+  "mcpServers": {
+    "uxarray": {
+      "command": "uxarray-mcp",
+      "args": ["serve"]
+    }
+  }
+}
+```
+
+**Any other MCP client**
+
+The server speaks stdio, so every client wants the same two facts — the
+command `uxarray-mcp` and the argument `serve`. `uxarray-mcp install-claude
+--print-only` prints the `mcpServers` JSON block that most clients accept
+verbatim.
 
 ### Step 4 — Sanity check
 
@@ -236,16 +320,27 @@ Intent-shaped tools, not raw UXarray bindings — all local by default:
 - `get_capabilities` — what can I do with this mesh?
 - `analyze_dataset` — deterministic first-look: inspect, validate, area, zonal mean, plots.
 - `run_analysis` — one operation at a time (gradient, curl, subset, remap, …).
-- `plot_dataset` — mesh, geographic, variable, or zonal-mean plots.
+  Remaps take `method`: `nearest_neighbor`, `inverse_distance_weighted` or
+  `bilinear` on UXarray's own engine, or `conservative`, `nnn`, `dnn`,
+  `average` on [YAC](https://dkrz-sw.gitlab-pages.dkrz.de/yac/) (equivalently
+  `backend="yac"` with `yac_method`). Only `conservative` preserves the field
+  integral, and it needs YAC importable where the remap runs — build it with
+  `scripts/build_yac_local.sh` on a laptop or `scripts/hpc_build_yac.py` on a
+  worker, and put its `site-packages` on `PYTHONPATH`.
+- `plot_dataset` — `plot_type` of `mesh`, `mesh_geo`, `variable`, or `zonal_mean`.
 - `run_workflow`, `resume_workflow`, `get_status`, `get_result`, `manage_session` —
   persisted sessions and multi-step workflows.
 
-Full schema: [docs/tools.md](docs/tools.md).
+Helper namespaces also appear in `tools/list`: `session/*`, `hpc/*`,
+`io/list_datasets`, `contract/*` and `prompt/*`.
+
+Full schema and every `run_analysis` parameter: [docs/tools.md](docs/tools.md).
 
 **Protocol version.** We do not implement MCP directly; servers are built
 through `toolregistry-server`, which depends on the `mcp` Python SDK. As of
 `toolregistry-server` 0.5.0 and `toolregistry` 0.16.0 the SDK cap is lifted, so
-we resolve `mcp` 2.1.1 and negotiate spec **`2026-07-28`** (stateless core,
+we resolve `mcp` 1.27 or 2.x (2.1.1 at the last lock) and negotiate spec
+**`2026-07-28`** (stateless core,
 cacheable list results, MRTR). 0.16.0 also widens the recognized content-block
 set to audio, `resource_link`, and embedded resources.
 
@@ -268,10 +363,13 @@ auditable and the server actively flags common scientific pitfalls:
   the tool that ran, timestamp, input arguments, `execution_venue`
   (`local` or `hpc:<endpoint>`), and the UXarray/Python versions used.
 - **Derivative unit convention is never hidden.** `gradient`, `curl`, and
-  `divergence` echo `scale_by_radius` in both the result and provenance, so a
-  unit-sphere result can never be mistaken for a physical (per-metre) one.
-  Gradient and curl default to physical scaling, matching UXarray; pass
-  `scale_by_radius=False` explicitly for unit-sphere output.
+  `divergence` echo `scale_by_radius` and a `radius_basis` block (the radius
+  used and whether it came from the grid or the caller), so a unit-sphere
+  result can never be mistaken for a physical (per-metre) one. Most grid files
+  declare no `sphere_radius`; pass `sphere_radius=6371000` (metres) to attach
+  Earth's. Without it, or with `scale_by_radius=False`, the call is refused
+  with `outcome="input_required"` until you pass `acknowledge`. The same
+  `sphere_radius` argument turns `calculate_area` from steradians into m².
 - **Vector-calculus sanity guard.** `curl`/`divergence` warn (without blocking)
   when the two inputs are the same field, or when neither carries a
   velocity/flux-like `units` attribute — the classic "vorticity from two random
@@ -293,12 +391,14 @@ auditable and the server actively flags common scientific pitfalls:
 
 | Command | Purpose |
 |---|---|
-| `uxarray-mcp serve` | Run the MCP server (used by your AI client) |
+| `uxarray-mcp serve` | Run the MCP server (used by your AI client); `--profile core\|deferred-full`, `--transport stdio\|sse\|http` — see [docs/serving.md](docs/serving.md) |
+| `uxarray-mcp openapi` | Print the OpenAPI document for the HTTP transport |
 | `uxarray-mcp setup` | Write a starter config |
 | `uxarray-mcp endpoints add NAME UUID` | Register a Globus Compute endpoint |
-| `uxarray-mcp endpoints list` | Show configured endpoints |
+| `uxarray-mcp endpoints list` / `remove NAME` | Show or drop configured endpoints |
+| `uxarray-mcp transfer setup` | Check and fix everything a Globus Transfer needs, including the browser login an MCP server cannot open |
 | `uxarray-mcp doctor` | Validate local + (optionally) remote setup |
-| `uxarray-mcp install-claude` | Merge or print the Claude Desktop config block |
+| `uxarray-mcp install-claude --config-path FILE` | Merge (or `--print-only`) the Claude Desktop config block |
 
 ---
 
@@ -339,7 +439,7 @@ see **[SECURITY.md](SECURITY.md)**.
 ## Development
 
 ```bash
-uv sync --extra hpc --extra docs --dev
+uv sync --extra hpc --extra transfer --extra docs
 uv run pre-commit run --all-files
 uv run pytest tests/ --ignore=tests/test_remote_agent.py
 uv run sphinx-build -b html docs docs/_build/html

{uxarray_mcp-0.3.0 → uxarray_mcp-2026.9.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "uxarray-mcp"
-version = "0.3.0"
+version = "2026.9.0"
 description = "MCP server for analyzing unstructured meshes with UXarray"
 readme = "README.md"
 keywords = [
@@ -25,11 +25,11 @@ requires-python = ">=3.11"
 dependencies = [
     "toolregistry-server[mcp]>=0.5.0",
     "toolregistry>=0.16.0",
-    "mcp>=1.24,<3",
+    "mcp>=1.27,<3",
     "holoviews>=1.19.0",
     "matplotlib>=3.9.0",
-    "pyyaml>=6.0",
-    "uxarray>=2026.8.0",
+    "pyyaml>=6.0.1",
+    "uxarray>=2026.9.0,<2026.10",
 ]
 
 [[project.authors]]
@@ -49,6 +49,7 @@ hpc = [
     "academy-py>=0.3.1",
     "globus-compute-sdk>=4.5.0",
 ]
+transfer = ["globus-cli>=3.30"]
 docs = [
     "sphinx>=7.0",
     "sphinx-book-theme>=1.1.0",
@@ -65,7 +66,9 @@ Tracker = "https://github.com/UXARRAY/uxarray-mcp-server/issues"
 
 [dependency-groups]
 dev = [
+    "jsonschema>=4.20.0",
     "mypy>=1.10.0",
+    "packaging>=23.0",
     "pre-commit>=4.3.0",
     "pytest>=9.0.2",
     "pytest-asyncio>=1.0.0",

{uxarray_mcp-0.3.0 → uxarray_mcp-2026.9.0}/pyproject.toml.orig

@@ -1,6 +1,6 @@
 [project]
 name = "uxarray-mcp"
-version = "0.3.0"
+version = "2026.9.0"
 description = "MCP server for analyzing unstructured meshes with UXarray"
 readme = "README.md"
 keywords = ["uxarray", "mcp", "unstructured grids", "scientific computing", "globus compute"]
@@ -45,11 +45,22 @@ dependencies = [
     # is_content_block_list about audio/resource_link/resource; on 0.15.0 those
     # blocks are silently stringified into one TextContent instead of failing.
     "toolregistry>=0.16.0",
-    "mcp>=1.24,<3",
+    # Both SDK generations are in range and register handlers differently:
+    # 1.27.2 has the per-method decorators and no `add_request_handler`, 2.1.1
+    # has `add_request_handler` and no decorators. `resources.py` registers
+    # through whichever is present, and `tests/test_artifact_resources.py`
+    # passes on both.
+    "mcp>=1.27,<3",
     "holoviews>=1.19.0",
     "matplotlib>=3.9.0",
-    "pyyaml>=6.0",
-    "uxarray>=2026.8.0",
+    # 6.0.1, not 6.0: PyYAML 6.0 ships no wheel for CPython 3.12 and its sdist
+    # cannot build against a modern Cython, failing with `'build_ext' object has
+    # no attribute 'cython_sources'`. Nothing hit it in normal CI because that
+    # resolves to the newest PyYAML; the weekly `lowest-direct` job installs the
+    # floor we actually declare, and had failed on it every run since
+    # 2026-08-10.
+    "pyyaml>=6.0.1",
+    "uxarray>=2026.9.0,<2026.10",
 ]
 
 [project.optional-dependencies]
@@ -60,6 +71,14 @@ hpc = [
     "academy-py>=0.3.1",
     "globus-compute-sdk>=4.5.0",
 ]
+# Moving files is separate from running code on them, and is done by running
+# the `globus` command-line client rather than by linking a second SDK. The CLI
+# owns the tokens, the consents and the session, so this package holds no auth
+# code at all -- which is the only reason a two-person project can offer data
+# movement without also signing up to track Globus's scope rules.
+transfer = [
+    "globus-cli>=3.30",
+]
 docs = [
     "sphinx>=7.0",
     "sphinx-book-theme>=1.1.0",
@@ -76,7 +95,16 @@ Tracker = "https://github.com/UXARRAY/uxarray-mcp-server/issues"
 
 [dependency-groups]
 dev = [
+    # Declared so the outputSchema conformance tests cannot silently skip.
+    # jsonschema arrives transitively today, which means a resolver change
+    # would turn the checks that hold every result to its published envelope
+    # into a green no-op.
+    "jsonschema>=4.20.0",
     "mypy>=1.10.0",
+    # Same reason as jsonschema: the suite-integrity check reads the declared
+    # uxarray floor out of the packaging metadata so it cannot drift from this
+    # file, and it needs a specifier parser to do that.
+    "packaging>=23.0",
     "pre-commit>=4.3.0",
     "pytest>=9.0.2",
     "pytest-asyncio>=1.0.0",

{uxarray_mcp-0.3.0 → uxarray_mcp-2026.9.0}/src/uxarray_mcp/__init__.py

@@ -3,4 +3,4 @@
 from uxarray_mcp.tools import inspect_mesh
 
 __all__ = ["inspect_mesh"]
-__version__ = "0.3.0"
+__version__ = "2026.9.0"

{uxarray_mcp-0.3.0 → uxarray_mcp-2026.9.0}/src/uxarray_mcp/app.py

@@ -23,6 +23,7 @@ from toolregistry_server.app import App
 
 from . import __version__
 from .registry import Profile
+from .resources import attach_artifact_resources
 
 if TYPE_CHECKING:
     from toolregistry import ToolRegistry
@@ -101,12 +102,16 @@ class UXarrayApp(App):
         name = kwargs.pop("name", identity.name)
         kwargs.pop("profile", None)
 
-        MCPAdapter(
+        adapter = MCPAdapter(
             route_table,
             name=name,
             list_tools_ttl_ms=LIST_TOOLS_TTL_MS,
             list_tools_cache_scope=LIST_TOOLS_CACHE_SCOPE,
-        ).run(**kwargs)
+        )
+        # Large figures go back as ``resource_link``, which is only honest if
+        # something answers ``resources/read`` for the URI (#14).
+        attach_artifact_resources(adapter.server)
+        adapter.run(**kwargs)
 
 
 # ---------------------------------------------------------------------------
@@ -122,16 +127,19 @@ def make_registry(*, profile: Profile = "core") -> ToolRegistry:
 def make_mcp_server(*, profile: Profile = "core"):
     """Build a configured MCP server ready for any transport.
 
-    Carries the same ``tools/list`` cache hints as :meth:`UXarrayApp.serve_mcp`
-    so tests and scripts exercise the surface the CLI actually serves.
+    Carries the same ``tools/list`` cache hints and the same artifact
+    resource handlers as :meth:`UXarrayApp.serve_mcp`, so tests and scripts
+    exercise the surface the CLI actually serves.
     """
     from toolregistry_server.adapters.mcp import route_table_to_mcp_server
     from toolregistry_server.route_table import RouteTable
 
     registry = make_registry(profile=profile)
-    return route_table_to_mcp_server(
-        RouteTable(registry),
-        "UXarray MCP",
-        list_tools_ttl_ms=LIST_TOOLS_TTL_MS,
-        list_tools_cache_scope=LIST_TOOLS_CACHE_SCOPE,
+    return attach_artifact_resources(
+        route_table_to_mcp_server(
+            RouteTable(registry),
+            "UXarray MCP",
+            list_tools_ttl_ms=LIST_TOOLS_TTL_MS,
+            list_tools_cache_scope=LIST_TOOLS_CACHE_SCOPE,
+        )
     )