inspector/core/mcp/uriTemplate.ts
cliffhall ebc18e1d56 fix: expand each expression independently, and reject invalid varspecs
Addresses Copilot's round-5 review on #2035 — eight suppressed comments,
all of them real.

The consequential one overturns behavior I copied from the SDK and
defended in two earlier rounds. RFC 6570 expands each expression
independently; it carries no cross-expression state. The SDK tracks
whether a query expression has already emitted and rewrites a later
`{?two}`'s `?` to `&`. That looks friendlier and is wrong, and its own
matcher says so: for `x{?one}{?two}` it expands to `x?one=1&two=2`, and
`match("x?one=1&two=2")` on that same template returns **null**, while
`match("x?one=1?two=2")` returns both variables. So the rewrite emitted a
URI the advertised template cannot match — #1919 one level up. A server
that wants a continuation advertises `{?one}{&two}`, which expands to `&`
and matches; producing that shape is the server's choice to declare, not
ours to infer.

Also:

- `{id:abc}` was silently treated as `{id}`, and a test codified it. RFC
  6570's max-length is `%x31-39 0*3DIGIT`, so `{id:}`, `{id:0}`,
  `{id:abc}` and `{id:10000}` are invalid templates rather than templates
  with an ignorable modifier — and the SDK's constructor accepts them all,
  so nothing else rejects them. Strict expansion now throws; lenient
  returns the raw template. Discovery stays lenient so the panel still
  renders rather than going blank.

- The TUI could submit `{a,b}` completely blank while the web panel
  blocked the same request: ink-form has no way to express "any one of
  these", so its members are left optional and the modal needs its own
  group check. Added, with tests — it was also below the coverage gate
  without them (functions 76.92%).

- Added the integration test the config never had: it resolves the
  checked-in config and drives all four reads over a real transport, so a
  misspelt preset fails there rather than only when someone runs the repro
  by hand.

- "All three clients" was false. The CLI has no template form and its
  `resources/read` passes the already-expanded `--uri` straight to
  `readResource`, so nothing here runs for it. Corrected in the README,
  AGENTS.md, and the module comment, and dropped a docblock still
  describing the SDK-delegation path removed last round.

Signed-off-by: cliffhall <cliff@futurescale.com>
2026-08-16 21:38:52 -04:00

482 lines
18 KiB
TypeScript

/**
* RFC 6570 URI Template parsing and expansion, shared by every client (#1919).
*
* This lives in `core/` rather than in a client so the web Resources form and
* the TUI cannot disagree about what a template means. Both derive their form
* fields from {@link templateVariables} and expand through
* {@link expandUriTemplate} — the web panel directly, the TUI via
* `InspectorClient.readResourceFromTemplate`.
*
* The **CLI is deliberately not a consumer**: it has no template form, and its
* `resources/read` passes the already-expanded `--uri` straight to
* `readResource` (see `clients/cli/src/handlers/run-method.ts`). Nothing here
* runs for it.
*
* ## Why this is not simply `new UriTemplate(t).expand(v)`
*
* The SDK's `UriTemplate` is still used, but only to *validate* a template —
* constructing it is what rejects an unclosed expression. Its expander is not,
* because it is incomplete in ways a form makes visible: a form has to *name*
* the variables it asks the user to fill in, so a parser that mangles a name
* produces a field nobody can use. Each of these was measured against the
* pinned SDK, not inferred:
*
* | Shape | SDK behavior |
* | ---------------- | --------------------------------------------------------- |
* | `{a,b}` | raw-joins the values — no encoding, operator prefix dropped |
* | `{;id}` | `;` is not in its operator list, so the variable is `;id` |
* | `{id:3}` | the prefix modifier is folded into the name, giving `id:3` |
* | `{+v}` / `{#v}` | `encodeURI` mangles reserved `[`/`]` and double-encodes `%` |
* | `{v}` | `encodeURIComponent` leaves the sub-delims `!'()*` bare |
*
* An earlier revision delegated the shapes the SDK got right and took over only
* the rest. That split is gone: once the encoders themselves differed, the two
* paths would have encoded the *same value* differently depending on whether
* the expression happened to carry a modifier. One expander, one set of rules.
*/
import { UriTemplate } from "@modelcontextprotocol/client";
/**
* The RFC 6570 operators.
*
* `;` is here but **not** in the SDK's own list, which is why `{;id}` parses
* there as a variable literally named `;id`. Order matters only in that each is
* a distinct single character; the first match wins.
*/
const OPERATORS = ["+", "#", ".", "/", ";", "?", "&"] as const;
/**
* Operators whose expansion omits cleanly when the variables in it are
* undefined — the whole expression, separator and all, simply disappears and
* what is left is still a well-formed URI naming a real (broader) resource.
*
* The remaining operators — `""` (simple) and `+` (reserved) — interpolate the
* value into the *middle* of the URI, so omitting one leaves an empty path
* segment rather than a shorter URI: measured against the pinned SDK,
* `file:///users/{userId}/profile` expands to `file:///users//profile` and
* `x://a/{+path}` to `x://a/`, both of which name a different resource. Those
* are the ones a form must require before allowing a read.
*
* `#` is in the omittable set on the same measurement: `x://a{#frag}` with no
* `frag` expands to exactly `x://a`. A fragment is optional by construction.
*/
const OMITTABLE_OPERATORS = new Set(["#", ".", "/", ";", "?", "&"]);
/** Operators that expand to `name=value` pairs rather than bare values. */
const NAMED_OPERATORS = new Set([";", "?", "&"]);
/** A single variable reference inside an expression, e.g. `id` or `id:3`. */
export interface VarSpec {
name: string;
/**
* The RFC 6570 prefix modifier (`{id:3}`), a maximum length in *characters*.
* Applied to the value before percent-encoding, per §3.2.1.
*/
maxLength?: number;
}
interface TemplateLiteral {
kind: "literal";
text: string;
}
interface TemplateExpression {
kind: "expression";
/** The expression including its braces, e.g. `{?topic}`. */
source: string;
/** The RFC 6570 operator, or `""` for a simple expression. */
operator: string;
/** The variable references, in order. */
varspecs: VarSpec[];
/** Bare variable names, `*` and any `:length` modifier stripped. */
names: string[];
/**
* True when a varspec carried a modifier that is not valid RFC 6570. Strict
* expansion rejects the template; discovery stays lenient so the panel can
* still render something rather than going blank.
*/
invalid: boolean;
}
export type TemplatePart = TemplateLiteral | TemplateExpression;
export interface TemplateVariable {
name: string;
/** The operator of the expression the variable was first seen in. */
operator: string;
/**
* True when this variable appears in at least one expression that cannot be
* omitted without changing the URI's structure — see
* {@link OMITTABLE_OPERATORS}. Drives the form's "Optional" marker.
*
* It does **not** mean "the user must fill this field in". Requiredness is a
* property of the *expression*: RFC 6570 drops undefined names from a
* multi-name expression, so `{a,b}` with only `a` filled expands to `a`'s
* value. Gate submission on {@link hasRequiredValues} over
* {@link requiredGroups}, never by testing this flag per variable.
*/
required: boolean;
}
/**
* RFC 6570's `max-length` production: `%x31-39 0*3DIGIT` — 1 to 9999, no
* leading zero. `{id:}`, `{id:0}`, `{id:abc}` and `{id:10000}` are all invalid
* *templates*, not templates with an ignorable modifier.
*/
const MAX_LENGTH_GRAMMAR = /^[1-9][0-9]{0,3}$/;
/**
* Parses one varspec (`id`, `id*`, `id:3`) into a name and optional prefix.
*
* Returns `null` for an empty varspec (a stray comma) and `"invalid"` for one
* whose modifier does not match the grammar. The two are distinguished because
* the first is ignorable and the second must fail the template: silently
* treating `{id:abc}` as `{id}` would send a URI that does not match what the
* server advertised, with nothing to alert anyone.
*/
function parseVarSpec(raw: string): VarSpec | null | "invalid" {
// The explode modifier is stripped rather than honored: it only changes how
// a list or map value is joined, and every value reaching this module is a
// single string.
const spec = raw.replace("*", "").trim();
if (spec.length === 0) return null;
const colon = spec.indexOf(":");
if (colon === -1) return { name: spec };
const name = spec.slice(0, colon);
const modifier = spec.slice(colon + 1);
if (name.length === 0 || !MAX_LENGTH_GRAMMAR.test(modifier)) return "invalid";
return { name, maxLength: Number(modifier) };
}
/**
* Splits a template into literal runs and expressions.
*
* An unclosed `{` yields a trailing literal, which is what makes the callers
* below degrade to "render no inputs, show the template verbatim" rather than
* throw at the user.
*/
export function parseUriTemplate(uriTemplate: string): TemplatePart[] {
const parts: TemplatePart[] = [];
let literal = "";
let i = 0;
while (i < uriTemplate.length) {
if (uriTemplate[i] !== "{") {
literal += uriTemplate[i];
i += 1;
continue;
}
const end = uriTemplate.indexOf("}", i);
if (end === -1) {
// Unclosed expression — the rest is not a template, treat it as text.
literal += uriTemplate.slice(i);
break;
}
if (literal) {
parts.push({ kind: "literal", text: literal });
literal = "";
}
const body = uriTemplate.slice(i + 1, end);
const operator = OPERATORS.find((op) => body.startsWith(op)) ?? "";
const parsed = body.slice(operator.length).split(",").map(parseVarSpec);
const varspecs = parsed.filter(
(spec): spec is VarSpec => spec !== null && spec !== "invalid",
);
parts.push({
kind: "expression",
source: uriTemplate.slice(i, end + 1),
operator,
varspecs,
names: varspecs.map((spec) => spec.name),
invalid: parsed.includes("invalid"),
});
i = end + 1;
}
if (literal) parts.push({ kind: "literal", text: literal });
return parts;
}
/**
* The variables a form should render an input for, in template order and
* deduplicated by name. A name appearing under more than one operator is
* required if *any* of its occurrences is.
*/
export function templateVariables(uriTemplate: string): TemplateVariable[] {
const byName = new Map<string, TemplateVariable>();
for (const part of parseUriTemplate(uriTemplate)) {
if (part.kind !== "expression") continue;
const required = !OMITTABLE_OPERATORS.has(part.operator);
for (const name of part.names) {
const existing = byName.get(name);
if (existing) {
existing.required = existing.required || required;
} else {
byName.set(name, { name, operator: part.operator, required });
}
}
}
return [...byName.values()];
}
/**
* The variable names of each expression that cannot be omitted, in template
* order — one entry per expression, not per variable.
*
* This is deliberately *not* folded onto {@link TemplateVariable}. A name can
* appear in several expressions with different operators, and each required
* expression has to be satisfied on its own: in `x{?a}{?b}{a,b}` the only
* required expression is `{a,b}`, which either `a` or `b` satisfies (the SDK
* expands that template with just `a` to `x?a=11`), while a per-variable model
* that kept only the first occurrence's group would mark both names required
* with singleton groups and refuse it. And in `{a,b}{a,c}` — two required
* expressions sharing `a` — filling `b` and `c` satisfies both, which no
* per-variable flag can express at all.
*/
export function requiredGroups(uriTemplate: string): string[][] {
const groups: string[][] = [];
for (const part of parseUriTemplate(uriTemplate)) {
if (part.kind !== "expression") continue;
if (OMITTABLE_OPERATORS.has(part.operator)) continue;
groups.push(part.names);
}
return groups;
}
/**
* Whether `values` supplies everything expansion structurally needs: every
* required expression has at least one of its names filled in.
*/
export function hasRequiredValues(
groups: string[][],
values: Record<string, string>,
): boolean {
return groups.every((names) =>
names.some((name) => (values[name] ?? "").length > 0),
);
}
/**
* Drops empty entries so an untouched optional field reads as *undefined*
* (the expression disappears) rather than as the empty string (which would
* expand to a valueless `?topic=`).
*/
export function definedValues(
values: Record<string, string>,
): Record<string, string> {
return Object.fromEntries(
Object.entries(values).filter(([, value]) => value.length > 0),
);
}
/**
* The characters RFC 6570 leaves alone under the `+` and `#` operators:
* RFC 3986 *unreserved* plus *reserved* (gen-delims and sub-delims).
*/
const ALLOW_RESERVED = /[^A-Za-z0-9\-._~:/?#[\]@!$&'()*+,;=]/gu;
/**
* The `allow-reserved` value encoding of RFC 6570 §3.2.1, used by `+` and `#`.
*
* The SDK reaches for `encodeURI` here, which is close but wrong twice over,
* and both cases corrupt the URI rather than merely over-escaping it:
*
* - It escapes `[` and `]`, which are *reserved* and must survive — so an IPv6
* literal `[::1]` becomes `%5B::1%5D`.
* - It escapes `%`, so an already-encoded value is double-encoded: `%2F`
* becomes `%252F`, and the server sees a literal "%2F" rather than a slash.
*
* The spec instead keeps existing pct-triplets intact, which is what the split
* below does: odd chunks are whole `%XX` triplets and pass through untouched,
* even chunks are scanned for anything outside the allowed set. A lone `%` is
* not a triplet, so it lands in an even chunk and is correctly encoded to
* `%25`. The `u` flag makes the class match by code point, so an astral
* character is handed to `encodeURIComponent` whole rather than as surrogates.
*/
function encodeAllowReserved(value: string): string {
return value
.split(/(%[0-9A-Fa-f]{2})/g)
.map((chunk, index) =>
index % 2 === 1
? chunk
: chunk.replace(ALLOW_RESERVED, (char) => encodeURIComponent(char)),
)
.join("");
}
/**
* Percent-encodes everything outside RFC 3986's *unreserved* set, which is what
* every operator except `+` and `#` calls for.
*
* `encodeURIComponent` alone is not that set: it leaves `!`, `'`, `(`, `)` and
* `*` unescaped. Those are sub-delims, not unreserved, so RFC 6570 requires
* them encoded for simple, label, path, matrix and query expansion. They are
* substituted afterwards rather than hand-rolled, so `encodeURIComponent` still
* does the UTF-8 work for everything else.
*/
const FORCE_ENCODED: Record<string, string> = {
"!": "%21",
"'": "%27",
"(": "%28",
")": "%29",
"*": "%2A",
};
function encodeUnreserved(value: string): string {
return encodeURIComponent(value).replace(
/[!'()*]/g,
(char) => FORCE_ENCODED[char],
);
}
/** Encodes one value for its operator: reserved characters survive `+` and `#`. */
function encodeValue(value: string, operator: string): string {
return operator === "+" || operator === "#"
? encodeAllowReserved(value)
: encodeUnreserved(value);
}
/**
* Applies a prefix modifier, then encodes.
*
* Truncation is by *code point* (`Array.from`), not by `slice`: RFC 6570 counts
* the prefix in characters, and `String.prototype.slice` counts UTF-16 code
* units, so it can cut an astral character in half and yield a lone surrogate.
*/
function renderValue(value: string, spec: VarSpec, operator: string): string {
const truncated =
spec.maxLength === undefined
? value
: Array.from(value).slice(0, spec.maxLength).join("");
return encodeValue(truncated, operator);
}
/** Expands one expression per RFC 6570. Returns "" if no name has a value. */
function expandExpression(
part: TemplateExpression,
values: Record<string, string>,
): string {
const present = part.varspecs.filter(
(spec) => values[spec.name] !== undefined,
);
if (present.length === 0) return "";
const { operator } = part;
if (NAMED_OPERATORS.has(operator)) {
const pairs = present.map(
(spec) =>
`${spec.name}=${renderValue(values[spec.name], spec, operator)}`,
);
// `;` repeats its separator per pair; `?`/`&` join with `&`.
return operator === ";"
? `;${pairs.join(";")}`
: `${operator}${pairs.join("&")}`;
}
const rendered = present.map((spec) =>
renderValue(values[spec.name], spec, operator),
);
switch (operator) {
case "#":
return `#${rendered.join(",")}`;
case ".":
return `.${rendered.join(".")}`;
case "/":
return `/${rendered.join("/")}`;
// "" and "+" — a bare comma-joined list, no prefix.
default:
return rendered.join(",");
}
}
/**
* Expands a whole parsed template.
*
* Each expression is expanded **independently**, which is what RFC 6570
* specifies — expansion carries no cross-expression state. The SDK instead
* tracks whether a query expression has already emitted and rewrites a later
* `{?two}`'s leading `?` to `&`. That looks friendlier and is wrong: measured
* against the pinned SDK, `x{?one}{?two}` expands to `x?one=1&two=2`, and
* `UriTemplate.match` on that same template *rejects* it — `match("x?one=1&two=2")`
* is `null`, while `match("x?one=1?two=2")` returns both variables. So the
* rewrite emits a URI the advertised template cannot match, which is precisely
* the failure #1919 is about, one level up.
*
* A server that wants a continuation advertises it: `{?one}{&two}` expands to
* `x?one=1&two=2` and matches. Producing that shape is the server's choice to
* declare, not ours to infer.
*/
export function expandParts(
parts: TemplatePart[],
values: Record<string, string>,
): string {
return parts
.map((part) =>
part.kind === "literal" ? part.text : expandExpression(part, values),
)
.join("");
}
/**
* Expands a template against the entered values per RFC 6570 — percent-encoding
* each value according to its operator, and omitting expressions whose
* variables were left blank. **Throws** on a template that is not valid.
*
* Every expression is expanded by {@link expandParts}; the SDK's `UriTemplate`
* is constructed only because that is what rejects an unclosed expression, and
* callers such as `readResourceFromTemplate` wrap the error it throws. Its own
* `expand` is deliberately unused — see the module comment for the five shapes
* it gets wrong.
*
* Its constructor is not a complete validator either: it accepts `{id:abc}`,
* whose modifier is not RFC 6570's `max-length` production. Treating that as a
* plain `{id}` would send a URI the server never advertised, so the owned
* parser's verdict is checked too.
*/
export function expandUriTemplateStrict(
uriTemplate: string,
values: Record<string, string>,
): string {
const defined = definedValues(values);
new UriTemplate(uriTemplate);
const parts = parseUriTemplate(uriTemplate);
const bad = parts.find((part) => part.kind === "expression" && part.invalid);
if (bad) {
throw new Error(
`Invalid RFC 6570 varspec in "${uriTemplate}": ${
bad.kind === "expression" ? bad.source : ""
} — a prefix modifier must be 1-9999 with no leading zero.`,
);
}
return expandParts(parts, defined);
}
/**
* {@link expandUriTemplateStrict}, but falling back to the raw template string
* instead of throwing.
*
* This is the form's variant: the template comes from the connected server, not
* from the user, so an invalid one is not something the user can fix from the
* panel — and what they already see in the URI preview is the raw template.
* Letting it throw would take out the whole panel on render; returning it
* unchanged sends the server a URI it rejects with a legible error instead.
* Call sites that need the failure (`readResourceFromTemplate`, which wraps it
* with the template name) use the strict variant.
*/
export function expandUriTemplate(
uriTemplate: string,
values: Record<string, string>,
): string {
try {
return expandUriTemplateStrict(uriTemplate, values);
} catch {
return uriTemplate;
}
}