diff --git a/README.md b/README.md
index c489bc7..517479a 100644
--- a/README.md
+++ b/README.md
@@ -86,6 +86,24 @@ repos:
       # - id: gmat-script-format-check  # or: check only, never write (CI)
 ```
 
+## Editor tooling
+
+[![VS Marketplace](https://img.shields.io/visual-studio-marketplace/v/astro-tools.gmat-script?label=VS%20Marketplace&color=311B92)](https://marketplace.visualstudio.com/items?itemName=astro-tools.gmat-script)
+[![Open VSX](https://img.shields.io/open-vsx/v/astro-tools/gmat-script?label=Open%20VSX&color=311B92)](https://open-vsx.org/extension/astro-tools/gmat-script)
+
+The same engine drives an editor experience — highlighting, hover docs, live diagnostics, completion,
+go-to-definition, an outline, and format-on-save.
+
+- **VS Code** — the **GMAT Script** extension, on the
+  [Visual Studio Marketplace](https://marketplace.visualstudio.com/items?itemName=astro-tools.gmat-script)
+  and [Open VSX](https://open-vsx.org/extension/astro-tools/gmat-script). Highlighting works on
+  install; the richer features come from the language server (`pip install "gmat-script[lsp]"`).
+- **Neovim, Emacs, Helix, and the rest** — a Language Server Protocol server (`gmat-script-lsp`,
+  from the `lsp` extra) backs any LSP-capable editor.
+
+See the [VS Code extension guide](https://astro-tools.github.io/gmat-script/vscode/) and the
+[language server guide](https://astro-tools.github.io/gmat-script/lsp/).
+
 ## The grammar surface
 
 The tree-sitter grammar parses GMAT scripts (`.script`) and GmatFunctions (`.gmf`) — the same
@@ -112,6 +130,11 @@ parsing is effectively version-independent — scripts from other releases parse
 *do* vary by release (valid field names, enums, defaults) belong to the linter and are scoped
 to R2026a.
 
+Those semantics live in a **field catalogue** reflected from R2026a (102 resource types, 2614
+fields) and shipped as data inside the wheel, so the linter and editor tooling need no GMAT install.
+It is version-pinned and provenance-stamped, selectable per release, and adding another GMAT version
+is a data file — not a code change. See [the field catalogue](https://astro-tools.github.io/gmat-script/catalogue/).
+
 ## What gmat-script is not
 
 - **Not a propagator or astrodynamics engine.** It reads and transforms script *text*; it computes
@@ -123,7 +146,8 @@ to R2026a.
 ## Documentation
 
 Full documentation — getting started, the grammar surface, the typed AST and editing guides, the
-formatter, the linter, the command-line tool, the error-reporting model, and the API reference — is at
+formatter, the linter and the field catalogue, the command-line tool, the language server and VS Code
+extension, the error-reporting model, and the API reference — is at
 **[astro-tools.github.io/gmat-script](https://astro-tools.github.io/gmat-script/)**.
 
 ## License
diff --git a/docs/api.md b/docs/api.md
index 9f8414f..e5e19d9 100644
--- a/docs/api.md
+++ b/docs/api.md
@@ -3,10 +3,12 @@
 The public surface of `gmat_script`. It is deliberately minimal and additive: the `parse` entry
 point and the `Tree` it returns, the `ErrorNode` / `Position` records that describe syntax errors,
 the typed `Script` overlay with its mutation API (`ObjectRef`, `RawValue`, `MutationError`), the
-canonical `format` pretty-printer, and the `lint` checker with its `Diagnostic` / `Severity` records.
-Each layer is re-exported here as it lands.
+canonical `format` pretty-printer, the `lint` checker with its `Diagnostic` / `Severity` records, and
+the `Catalog` field catalogue (`load_catalog`, `FieldSpec`, `TypeSpec`). Each layer is re-exported
+here as it lands.
 
 For a worked introduction see [Getting started](getting-started.md); for the error model see
-[Error reporting](errors.md); for the linter see [Linting](lint.md).
+[Error reporting](errors.md); for the linter see [Linting](lint.md); for the catalogue see
+[The field catalogue](catalogue.md).
 
 ::: gmat_script
diff --git a/docs/catalogue.md b/docs/catalogue.md
new file mode 100644
index 0000000..0f8743a
--- /dev/null
+++ b/docs/catalogue.md
@@ -0,0 +1,120 @@
+# The field catalogue
+
+The field catalogue is the knowledge base behind everything that reasons about GMAT *semantics*: the
+[linter](lint.md), and the [language server](lsp.md)'s hover docs and completion. It records, for
+every GMAT resource type, the fields that type defines — each field's type, allowed enumeration,
+reference target, default, and unit — so a tool can tell that `Spacecraft.SMA` wants a number, that
+`ImpulsiveBurn.Axes` is one of four values, or that `Spacecraft.Tanks` points at a `FuelTank`,
+without running GMAT.
+
+It is shipped as **data**, not derived at runtime. The wheel carries a precompiled JSON catalogue;
+loading it imports no GMAT and touches no `gmatpy`.
+
+## Where it comes from — and the GMAT-free guarantee
+
+The catalogue is reflected from a real GMAT install at build time, by a single generator, and then
+frozen into the package. Two halves, deliberately kept apart:
+
+- **Generation (build/CI time, GMAT required).** `gmat_script.tools.gen_catalog` is the *only*
+  GMAT-touching code in the project. It imports `gmatpy`, walks GMAT's object factory and each
+  object's parameter metadata, normalises the types, and writes
+  `gmat_script/data/fields-<version>.json`.
+- **Loading (runtime, no GMAT).** `gmat_script.catalog` reads that JSON and answers queries. It
+  **never imports `gmatpy`** — so a plain `pip install gmat-script` carries the full catalogue with
+  no GMAT install anywhere.
+
+This split is the GMAT-free guarantee in practice: GMAT's knowledge is captured once, offline, and
+travels with the package as a data file. (See the [design decisions](design/decisions.md).)
+
+The shipped catalogue is reflected from **GMAT R2026a** and records its 102 resource types and 2614
+fields.
+
+## Reading the catalogue
+
+`load_catalog()` returns a cached, queryable `Catalog`. It and its records are part of the public
+API:
+
+```python
+from gmat_script import load_catalog
+
+cat = load_catalog()
+
+cat.gmat_version            # 'R2026a' — the release this was reflected from
+cat.generated               # '2026-06-08' — the ISO date it was generated
+
+cat.has_type("Spacecraft")          # True
+cat.field_type("Spacecraft", "SMA") # 'real'
+cat.enum_values("ImpulsiveBurn", "Axes")
+# ('VNB', 'LVLH', 'MJ2000Eq', 'SpacecraftBody')
+cat.ref_target("Spacecraft", "Tanks")   # 'FuelTank'
+```
+
+A type name that GMAT spells differently in scripts resolves through an alias (`Propagator` ->
+`PropSetup`, `ODEModel` -> `ForceModel`), so `cat.has_type("Propagator")` is `True`. Any query for an
+unknown type or field returns `None` (or `[]`) rather than raising — the linter and editor leans on
+this to degrade to "no finding" wherever the catalogue is silent.
+
+### What a field carries
+
+`cat.field(type, name)` returns a `FieldSpec`:
+
+| Attribute | Meaning |
+|-----------|---------|
+| `type` | The normalised catalogue type: `real`, `integer`, `string`, `bool`, `enum`, `object`, `object_array`, `string_array`, `real_array`, `matrix`, `filename`, `on_off`, `color`, `gmat_time`, … |
+| `gmat_type` | GMAT's own raw type label, kept verbatim (e.g. `Real`, `Rmatrix`). |
+| `read_only` | Whether GMAT exposes the field as output-only (not settable from a script). |
+| `allowed` | The enum's allowed values, where GMAT exposes them — else `None`. |
+| `ref_target` | The target GMAT type for an object-reference field — else `None`. |
+| `default` | The default for a settable scalar field — else `None`. |
+| `unit` | The field's unit, where GMAT reports one — else `None`. |
+
+`cat.type_spec(type)` returns a `TypeSpec` carrying the type's GMAT object-type `category` and its
+`fields` mapping. The lower-level `Catalog.load(target_version=...)` builds an uncached catalogue;
+`load_catalog` is the convenience entry point most consumers want.
+
+## Versioning
+
+The semantics a script must satisfy — valid field names, enums, defaults — vary by GMAT release, so
+the catalogue is **version-pinned and provenance-stamped**: every catalogue carries the GMAT version
+it was reflected from and the date it was generated. The grammar, by contrast, never enumerates types
+or keywords and is effectively version-independent; the catalogue is the one version-coupled artifact.
+
+`load_catalog()` defaults to the newest shipped catalogue. Pass `target_version` to pin a specific
+release:
+
+```python
+load_catalog("R2026a")        # an explicit release
+load_catalog()                # the newest available (currently R2026a)
+```
+
+The linter exposes the same selector — `lint(source, target_version="R2026a")` — and the language
+server uses the default. Requesting a version that is not shipped raises `ValueError` listing what is
+available.
+
+Supporting another GMAT release is **additive**: a new `fields-<version>.json` is dropped into
+`gmat_script/data/`, and `load_catalog()`'s "newest" default and the `target_version` selector pick
+it up with no code change. There is no per-version code fork.
+
+## Regenerating the catalogue (the version-bump process)
+
+Regenerating is only needed when targeting a new GMAT release or picking up a point-release metadata
+change. It needs a real GMAT install — `gmatpy` is imported from it directly and is never
+pip-installed.
+
+```console
+# Regenerate and overwrite the shipped file, against a GMAT install:
+$ python -m gmat_script.tools.gen_catalog
+# Regenerate in memory and fail on any drift from the committed catalogue (the CI check):
+$ python -m gmat_script.tools.gen_catalog --check
+```
+
+The generator locates GMAT from `--gmat-root`, else the `GMAT_ROOT` environment variable, else a set
+of platform-standard install paths. A dedicated CI job runs the `--check` form on a schedule (and on
+demand) against a freshly provisioned GMAT install, so a catalogue that drifts from the current GMAT
+metadata is surfaced rather than silently rotting. The generation date is ignored in the drift
+comparison, so re-running on a different day is not spurious drift.
+
+To bump to a new release: regenerate against that GMAT version (so the file is written as
+`fields-<new-version>.json`), commit the new data file alongside the existing one, and the loader's
+"newest" default begins serving it. Keeping the prior file lets callers pin the older release through
+`target_version`.
diff --git a/docs/index.md b/docs/index.md
index e0f2f9b..63ed880 100644
--- a/docs/index.md
+++ b/docs/index.md
@@ -12,9 +12,9 @@ tree.has_errors      # False
 tree.to_source()     # round-trips byte-for-byte to the input
 ```
 
-gmat-script ships the parser, a typed AST with a mutation API, and a canonical formatter, with a
-`gmat-script` command-line tool over the same engine. The linter and editor tooling build on top of
-the same tree as they land.
+gmat-script ships the parser, a typed AST with a mutation API, a canonical formatter, and a static
+linter, with a `gmat-script` command-line tool over the same engine — plus a language server and a
+VS Code extension that bring it all to your editor.
 
 ## What it is
 
@@ -22,8 +22,7 @@ the same tree as they land.
   full R2026a sample corpus and re-emits it byte-for-byte.
 - A Python library that loads that grammar from a **vendored, precompiled** binding — so
   `pip install gmat-script` needs no C or Node toolchain, and never GMAT.
-- A `gmat-script` command-line tool that parses and formats scripts from the shell or CI (with
-  linting to follow as it lands).
+- A `gmat-script` command-line tool that parses, formats, and lints scripts from the shell or CI.
 
 ## What it is not
 
@@ -41,7 +40,11 @@ the same tree as they land.
 - **[Typed AST](typed-ast.md)** — typed resources and dict-like field access over the tree.
 - **[Editing](editing.md)** — set fields, rename resources, and splice commands.
 - **[Formatter](formatting.md)** — canonical, idempotent re-emission.
-- **[CLI](cli.md)** — the `parse` syntax gate and the `format` command.
+- **[Linter](lint.md)** — structural checks against the bundled field catalogue.
+- **[Field catalogue](catalogue.md)** — the version-pinned knowledge base behind the linter and editor.
+- **[CLI](cli.md)** — the `parse` syntax gate and the `format` and `lint` commands.
+- **[Language server](lsp.md)** — diagnostics, hover, and completion in any LSP editor.
+- **[VS Code extension](vscode.md)** — highlighting, diagnostics, and format-on-save in VS Code.
 - **[Error reporting](errors.md)** — how malformed input is surfaced.
 - **[API reference](api.md)** — the public Python surface.
 - **[Design decisions](design/decisions.md)** — the grammar scope, CST node taxonomy, and the
diff --git a/docs/lsp.md b/docs/lsp.md
index 95e7cc6..0297e0d 100644
--- a/docs/lsp.md
+++ b/docs/lsp.md
@@ -69,12 +69,9 @@ vim.api.nvim_create_autocmd("FileType", {
 
 ### VS Code
 
-Install the [**GMAT Script**](https://marketplace.visualstudio.com/items?itemName=astro-tools.gmat-script)
-extension (also on [Open VSX](https://open-vsx.org/extension/astro-tools/gmat-script)). It bundles a
-client that launches this server for you, and adds standalone syntax highlighting that works even
-before the server is installed. Add the server with `pip install "gmat-script[lsp]"` — the extension
-picks up `gmat-script-lsp` from your `PATH`, or point `gmatScript.server.pythonPath` at the Python
-environment that has it. Format-on-save is enabled for GMAT files by default.
+The [**GMAT Script**](vscode.md) extension bundles a client that launches this server for you, plus
+standalone syntax highlighting that works before the server is installed. See the [VS Code extension
+guide](vscode.md) for install (Marketplace / Open VSX), features, format-on-save, and settings.
 
 ## How it fits together
 
diff --git a/docs/vscode.md b/docs/vscode.md
new file mode 100644
index 0000000..def540a
--- /dev/null
+++ b/docs/vscode.md
@@ -0,0 +1,71 @@
+# The VS Code extension
+
+**GMAT Script** brings the grammar, linter, and [language server](lsp.md) to Visual Studio Code:
+syntax highlighting, hover docs, live diagnostics, completion, go-to-definition, an outline, and
+format-on-save for `.script` and `.gmf` files.
+
+## Installing
+
+Install **GMAT Script** from either marketplace:
+
+- [Visual Studio Marketplace](https://marketplace.visualstudio.com/items?itemName=astro-tools.gmat-script)
+  — search *GMAT Script* in the Extensions view, or run
+  `code --install-extension astro-tools.gmat-script`.
+- [Open VSX](https://open-vsx.org/extension/astro-tools/gmat-script) — for VSCodium, Cursor, Gitpod,
+  and other Open VSX–based editors.
+
+The extension bundles **syntax highlighting** as a TextMate grammar, so colouring works the moment
+it is installed — no Python, no further setup.
+
+Everything richer — hover, diagnostics, completion, definition, references, the outline, and
+formatting — is served by the `gmat-script` language server, which runs on Python. Install it into a
+Python environment the extension can reach:
+
+```console
+$ pip install "gmat-script[lsp]"
+```
+
+This puts a `gmat-script-lsp` command on your `PATH`, which the extension launches automatically. If
+it is not found, highlighting still works and the extension points you here. To use a specific
+interpreter or virtual environment, set `gmatScript.server.pythonPath` (see
+[Settings](#settings)).
+
+## Features
+
+| Feature | Needs the server? |
+|---------|:-----------------:|
+| Syntax highlighting — resources, commands, fields, control flow, solver blocks, GmatFunction headers | no |
+| Hover — a field's type, default, allowed values, units, and reference target | yes |
+| Live diagnostics as you type — syntax errors plus linter findings | yes |
+| Completion — resource names, the fields valid for the resource under the cursor, and enum values | yes |
+| Go to definition / find all references for resources | yes |
+| Document outline — every `Create`'d resource and GmatFunction header | yes |
+| Format on save — canonical, idempotent re-emission | yes |
+
+The server-backed features are exactly those of the [language server](lsp.md); the extension is the
+VS Code client that launches and talks to it. A half-typed buffer never breaks them — the
+error-recovering parse keeps hover and completion working while you edit.
+
+## Format on save
+
+Format-on-save is **enabled for GMAT files by default**: the extension sets `editor.formatOnSave`
+for the `gmat` language, so saving a `.script` or `.gmf` rewrites it into canonical form via the same
+formatter the [CLI](cli.md) and library use. Turn it off per-language if you prefer:
+
+```jsonc
+"[gmat]": { "editor.formatOnSave": false }
+```
+
+## Settings
+
+| Setting | Default | Description |
+|---|---|---|
+| `gmatScript.server.path` | `gmat-script-lsp` | Command used to launch the language server. Set an absolute path if it is not on your `PATH`. |
+| `gmatScript.server.pythonPath` | _(empty)_ | A Python interpreter with `gmat-script[lsp]` installed; when set, the server runs as `<python> -m gmat_script.lsp` and takes precedence over `server.path`. Point it at a virtual environment. |
+| `gmatScript.server.args` | `[]` | Extra arguments passed to the server on launch. |
+| `gmatScript.trace.server` | `off` | Trace JSON-RPC traffic to the output channel (`off` / `messages` / `verbose`) when debugging. |
+
+## Other editors
+
+The same server powers any LSP-capable editor — Neovim, Emacs, Helix, and the rest. See the
+[language server guide](lsp.md) for editor-agnostic setup.
diff --git a/editors/vscode/images/README.md b/editors/vscode/images/README.md
index a4fe859..2cda16d 100644
--- a/editors/vscode/images/README.md
+++ b/editors/vscode/images/README.md
@@ -1,8 +1,48 @@
 # Marketing assets
 
 `demo.gif` — a short animated capture of the extension in action (highlighting + live diagnostics +
-hover on a stock GMAT sample), shown at the top of the extension's marketplace listing.
+hover + format-on-save), shown at the top of the extension's marketplace listing and referenced from
+`../README.md` (the reference is commented out until the file exists).
 
-To add it: record a `.script` editing session in VS Code (type into a sample, trigger a diagnostic,
-hover a field), export an optimized GIF (≈800 px wide, a few seconds, looping), save it here as
-`demo.gif`, and uncomment the image reference in `../README.md`.
+## Recording `demo.gif`
+
+A repeatable ~8-second capture. Keep it small and looping.
+
+**Setup**
+
+1. Install the extension (run the *Extension Development Host* from this folder, or install the
+   packaged `.vsix`) and `pip install "gmat-script[lsp]"` into the interpreter VS Code uses, so the
+   language server is live.
+2. Use a clean window: dark theme, no minimap, font size ~16, a single editor column ~80 columns
+   wide. Hide the sidebar and status-bar clutter.
+3. Create a new file `demo.script` and leave it empty.
+
+**The take** (type at a natural pace; let each step settle for ~1 s)
+
+1. Type the snippet below. Highlighting colours resources, fields, the burn axis, and commands as you
+   go:
+
+   ```text
+   Create Spacecraft Sat
+   Sat.SMA = 7000
+
+   Create ImpulsiveBurn TOI
+   TOI.Axes = VNB
+
+   BeginMissionSequence
+   Maneuver TOI(Sat)
+   ```
+
+2. Change `Sat.SMA = 7000` to `Sat.SMA = 'high'`. A red squiggle appears under `'high'`; hover it to
+   show the `type-mismatch` diagnostic ("field 'SMA' expects a number, got a quoted string").
+3. Hover `Axes` to show the field doc card (type + allowed values `VNB, LVLH, MJ2000Eq,
+   SpacecraftBody`).
+4. Fix `Sat.SMA` back to `7000`; the squiggle clears.
+5. Mangle the spacing (e.g. `GMAT Sat.SMA=7000;`) and press **Save** — format-on-save snaps it back to
+   canonical form.
+
+**Export**
+
+Capture the editor region only, export an optimized looping GIF (≈800 px wide, a few seconds), save
+it here as `demo.gif`, and uncomment the `![GMAT Script in VS Code](images/demo.gif)` line in
+`../README.md`.
diff --git a/examples/README.md b/examples/README.md
index 2e4d898..e961c0e 100644
--- a/examples/README.md
+++ b/examples/README.md
@@ -13,8 +13,10 @@ python examples/edit_field.py
 | [`edit_field.py`](edit_field.py) | Read a resource field, then set it — `script.spacecraft["Sat"]["SMA"] = 8000`. |
 | [`rename_resource.py`](rename_resource.py) | Rename a resource with and without rewriting its references. |
 | [`format_in_place.py`](format_in_place.py) | Format a messy script into canonical form, written back to the file. |
+| [`lint_script.py`](lint_script.py) | Lint a flawed script — type, field, reference-target, and enum findings — then show it clean. |
 
 For the concepts behind these, see the documentation:
 [the typed AST](https://astro-tools.github.io/gmat-script/typed-ast/),
-[editing](https://astro-tools.github.io/gmat-script/editing/), and
-[the formatter](https://astro-tools.github.io/gmat-script/formatting/).
+[editing](https://astro-tools.github.io/gmat-script/editing/),
+[the formatter](https://astro-tools.github.io/gmat-script/formatting/), and
+[the linter](https://astro-tools.github.io/gmat-script/lint/).
diff --git a/examples/lint_script.py b/examples/lint_script.py
new file mode 100644
index 0000000..5e91182
--- /dev/null
+++ b/examples/lint_script.py
@@ -0,0 +1,65 @@
+"""Lint a GMAT script: print the diagnostics for a flawed script, then show it clean.
+
+Run with:  python examples/lint_script.py
+
+The linter checks a script's *structure* against the bundled field catalogue — unknown types and
+fields, type / enum / reference-target mismatches, duplicate names, and unused or undeclared
+references. It never runs the mission and needs no GMAT install.
+"""
+
+from __future__ import annotations
+
+from gmat_script import lint
+
+# A syntactically valid script with several seeded *structural* problems the linter catches.
+FLAWED = """\
+Create Spacecraft Sat
+Sat.SMA = 'high'
+Sat.Naem = 7000
+
+Create Thruster Thr
+Sat.Tanks = {Thr}
+
+Create ImpulsiveBurn TOI
+TOI.Axes = Sideways
+
+BeginMissionSequence
+Maneuver TOI(Sat)
+"""
+
+# The same mission with each finding addressed: a numeric SMA, the misspelled field corrected, a
+# FuelTank in the tank slot, and a valid burn axis. The linter reports nothing.
+CLEAN = """\
+Create Spacecraft Sat
+Sat.SMA = 7000
+Sat.Id = 'SAT-1'
+
+Create FuelTank Tank
+Sat.Tanks = {Tank}
+
+Create ImpulsiveBurn TOI
+TOI.Axes = VNB
+
+BeginMissionSequence
+Maneuver TOI(Sat)
+"""
+
+
+def report(label: str, source: str) -> None:
+    print(f"--- {label} ---")
+    diagnostics = lint(source)
+    if not diagnostics:
+        print("no findings")
+        return
+    for d in diagnostics:
+        print(f"{d.start.line}:{d.start.column} {d.severity} {d.rule}: {d.message}")
+
+
+def main() -> None:
+    report("flawed script", FLAWED)
+    print()
+    report("cleaned script", CLEAN)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/mkdocs.yml b/mkdocs.yml
index 51add51..4ac10c8 100644
--- a/mkdocs.yml
+++ b/mkdocs.yml
@@ -40,8 +40,11 @@ nav:
   - Editing: editing.md
   - Formatter: formatting.md
   - Linter: lint.md
+  - Field catalogue: catalogue.md
   - CLI: cli.md
-  - Language server: lsp.md
+  - Editor tooling:
+      - Language server: lsp.md
+      - VS Code extension: vscode.md
   - Error reporting: errors.md
   - API reference: api.md
   - Design: