toolmenu 0.12.0 → 0.14.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (63) hide show
  1. package/README.md +84 -23
  2. package/dist/args.d.ts +31 -5
  3. package/dist/args.js +91 -14
  4. package/dist/auth.d.ts +4 -2
  5. package/dist/auth.js +59 -21
  6. package/dist/auto.d.ts +80 -6
  7. package/dist/auto.js +255 -19
  8. package/dist/breakdown.d.ts +2 -0
  9. package/dist/breakdown.js +64 -19
  10. package/dist/catalog.d.ts +87 -10
  11. package/dist/catalog.js +415 -58
  12. package/dist/cli.js +170 -185
  13. package/dist/close.d.ts +8 -0
  14. package/dist/close.js +18 -0
  15. package/dist/config.d.ts +7 -1
  16. package/dist/config.js +17 -0
  17. package/dist/connect.d.ts +14 -0
  18. package/dist/connect.js +150 -16
  19. package/dist/diff.d.ts +23 -1
  20. package/dist/diff.js +574 -120
  21. package/dist/difference.d.ts +2 -0
  22. package/dist/difference.js +13 -12
  23. package/dist/explain.d.ts +113 -0
  24. package/dist/explain.js +461 -0
  25. package/dist/failures.d.ts +15 -0
  26. package/dist/failures.js +59 -2
  27. package/dist/help.d.ts +11 -0
  28. package/dist/help.js +205 -0
  29. package/dist/history.d.ts +13 -0
  30. package/dist/history.js +86 -6
  31. package/dist/init.d.ts +93 -2
  32. package/dist/init.js +508 -21
  33. package/dist/menu.js +10 -2
  34. package/dist/options.d.ts +141 -0
  35. package/dist/options.js +50 -0
  36. package/dist/report.d.ts +2 -2
  37. package/dist/report.js +305 -101
  38. package/dist/rules/description.d.ts +7 -2
  39. package/dist/rules/description.js +86 -42
  40. package/dist/rules/determinism.d.ts +4 -1
  41. package/dist/rules/determinism.js +175 -11
  42. package/dist/rules/gated.d.ts +7 -0
  43. package/dist/rules/gated.js +77 -0
  44. package/dist/rules/ids.d.ts +1 -1
  45. package/dist/rules/ids.js +9 -4
  46. package/dist/rules/index.d.ts +6 -0
  47. package/dist/rules/index.js +18 -6
  48. package/dist/rules/naming.d.ts +1 -1
  49. package/dist/rules/naming.js +91 -21
  50. package/dist/rules/rule.d.ts +25 -0
  51. package/dist/rules/rule.js +25 -1
  52. package/dist/rules/spec.d.ts +6 -0
  53. package/dist/rules/spec.js +56 -9
  54. package/dist/rules/write.d.ts +1 -1
  55. package/dist/rules/write.js +64 -14
  56. package/dist/session.d.ts +34 -9
  57. package/dist/session.js +525 -94
  58. package/dist/snapshot.d.ts +8 -2
  59. package/dist/snapshot.js +33 -2
  60. package/dist/types.d.ts +21 -0
  61. package/dist/words.d.ts +5 -0
  62. package/dist/words.js +14 -0
  63. package/package.json +2 -2
package/README.md CHANGED
@@ -29,7 +29,7 @@ No LLM anywhere: the same inputs give the same answer, so it can sit in CI.
29
29
  (`history` installs old versions with today's dependencies, so it records what they
30
30
  serve now, which can differ from what they shipped with: FINDINGS F4.)
31
31
 
32
- > **Status: 0.12, early.** Spec in [docs/SPEC.md](https://github.com/niksa90/toolmenu/blob/main/docs/SPEC.md). What it has found on
32
+ > **Status: 0.14, early.** Spec in [docs/SPEC.md](https://github.com/niksa90/toolmenu/blob/main/docs/SPEC.md). What it has found on
33
33
  > real servers: [docs/FINDINGS.md](https://github.com/niksa90/toolmenu/blob/main/docs/FINDINGS.md).
34
34
 
35
35
  ## Start here
@@ -50,6 +50,14 @@ Every value you pass with `--env` or `--header` goes into the workflow as
50
50
  `${{ secrets.… }}`, never as text (a name like `DATABASE_URL` doesn't say it holds a
51
51
  password); `init` says which secrets to add. It never overwrites a file.
52
52
 
53
+ Run it anywhere in the repository. `.github/workflows/toolmenu.yml` goes to the git
54
+ root, and the server starts in CI from the folder you ran `init` in, which is also
55
+ where `menu.json` is saved. That's how a monorepo package works: run `init` in
56
+ `packages/server`. Paths to files in the repository are made relative, so they work on
57
+ the runner. `init` lists anything that won't work there (a file outside the
58
+ repository, a `localhost` URL, a virtualenv the runner doesn't have), each with what
59
+ to do.
60
+
53
61
  By hand:
54
62
 
55
63
  ```sh
@@ -75,6 +83,11 @@ dynamic registration needs a pre-registered app: `--client-id` (and
75
83
  Both protocol generations are supported: 2026-07-28 (`server/discover`, stateless)
76
84
  and the 2025 `initialize` handshake, through the official TypeScript SDK.
77
85
 
86
+ Every finding and error says what's wrong, where, what toolmenu saw, and ends in a
87
+ `→ Next:` step. Findings marked `· unsure` are heuristics (names and words): they say
88
+ what was seen, not what it means. Every command has its own help:
89
+ `toolmenu history --help`.
90
+
78
91
  ## In CI: one PR comment
79
92
 
80
93
  ```yaml
@@ -91,7 +104,7 @@ jobs:
91
104
  - uses: actions/setup-node@v7
92
105
  with: { node-version: 22 }
93
106
  - run: npm ci && npm run build
94
- - uses: niksa90/toolmenu@v0.12.0
107
+ - uses: niksa90/toolmenu@v0.14.0
95
108
  with:
96
109
  command: node dist/server.js
97
110
  baseline: menu.json # your committed snapshot
@@ -166,7 +179,9 @@ serves, and 23 issue and pull-request tools that only the dynamic mode has.
166
179
  | `session/unannounced` | warn | The menu changed without `notifications/tools/list_changed`, although the server declared `listChanged` |
167
180
  | `session/untested` | warn, error if no call got through | Calls that failed before reaching the tool: authentication, something missing on this machine (no Chrome), the network. One finding for the run, not one per step, so an expired CI secret doesn't pass a run that tested nothing |
168
181
  | `session/refused`, `session/step-failed` | error | A write the scenario didn't allow, or a call that failed for another reason |
169
- | `session/tool-error` | warn | A tool that answered with an error for another reason (`isError`): the run tested less than it looks |
182
+ | `session/tool-error` | warn | A tool that answered with an error for another reason (`isError`): the run tested less than it looks. The same error at several steps is one finding that names them |
183
+ | `session/known-variance` | info | A mid-session change only in values that already vary from one tools/list to the next: same cause as the finding it names, not counted again |
184
+ | `session/assumed-read-only` | info | Tools called on your word (`--assume-read-only`, `assume_read_only`) although the server doesn't mark them readOnlyHint |
170
185
  | `session/unlock-coverage` | warn | An unlock whose enum lists its values, called with only some of them: the tools behind the rest were never seen, so a baseline from the session misses them. With `--union-out` or `--auto`, an unlock never called at all too, with why `--auto` skipped it |
171
186
  | `session/nothing-called` | warn | `--auto` called no tools (all open-world, missing values, or not read-only): the run only listed the menu |
172
187
  | `session/session-lost` | error | The server ended the session after toolmenu opened a second connection with the same credentials (servers with one session per client). The run stops there; rerun with `--processes 1` |
@@ -177,10 +192,17 @@ serves, and 23 issue and pull-request tools that only the dynamic mode has.
177
192
  whose required arguments the schema itself gives (a `default`, `examples`, an `enum`,
178
193
  a type or format), cheapest first, then every value of each read-only unlock (outside
179
194
  `--max-calls`, and the first value twice: a second identical unlock should change
180
- nothing), then the first call again. It never guesses an ID:
181
- a tool that needs `owner`, `repo` or an issue key is skipped, and the report says
182
- which values were missing (`--save-scenario auto.yml` writes the steps, with the
183
- skipped tools commented out, to fill in). Tools marked `openWorldHint: true` (web
195
+ nothing), then the first call again. It never guesses an ID: a tool that needs
196
+ `owner`, `repo` or an issue key is skipped, and the report names the flag that fills
197
+ it: `--value repo_path=/src/app` fills that parameter on every tool that requires it,
198
+ `--value get_issue.issue_key=ABC-1` on one tool, `--values-file values.yml` takes a map
199
+ of either. Servers that don't mark their tools `readOnlyHint` (DeepWiki, every
200
+ 2024-11-05 server) get nothing called unless you vouch for them by name:
201
+ `--assume-read-only read_wiki_structure,ask_wiki_question`. Only exact names are
202
+ accepted: a pattern would also cover tools a later release adds. A tool marked or
203
+ named as a write is never called, and the report says which tools ran on your word.
204
+ `--save-scenario auto.yml` writes the steps, with the skipped tools commented out, for
205
+ you to fill in. Tools marked `openWorldHint: true` (web
184
206
  search, fetch, scraping, which can cost API credits) are called only with
185
207
  `--open-world`, except unlocks: changing the menu is what a session watches, and an
186
208
  unlock spends no search credits, so it runs even on a server that marks every tool
@@ -200,11 +222,16 @@ npx toolmenu diff menu.json new-menu.json
200
222
  ```
201
223
  $ npx toolmenu diff --release 2026.1.14..2026.8.31 fs-2026.1.14.json fs-2026.8.31.json
202
224
  toolmenu diff secure-filesystem-server 2026.1.14 → 2026.8.31
203
- 14 → 14 tools · ~1,640 → ~1,664 tokens (+24, estimate): this release adds ~24 tokens to every conversation that loads the menu
204
- 1 breaking · 0 minor · 15 notice · suggested bump: major · actual: not checked (calendar version)
225
+ changes 1 breaking (1 tool) · 0 minor · 3 notice (13 tools)
226
+ version suggested bump: major · not checked (calendar version)
227
+ tokens ~1,640 → ~1,664 (+24, estimate): this release adds ~24 tokens to every conversation that loads the menu
228
+ tools 14 → 14
205
229
 
206
230
  ERROR diff/safety-hint
207
- move_file was additive-only and is now destructive.
231
+ `move_file` was marked additive-only and is now destructive (destructiveHint false → true; unset means destructive). Clients that auto-approve additive tools may now ask first.
232
+ → Next: If the tool still never deletes or overwrites, set destructiveHint: false; if it does now, say so in the release notes.
233
+ INFO diff/annotations
234
+ Annotations changed the same way (openWorldHint (unset) → false) in 13 of 14 tools: read_file, read_text_file, read_media_file, read_multiple_files, write_file, edit_file, … (+7).
208
235
  …
209
236
  ```
210
237
 
@@ -215,9 +242,9 @@ the way OpenAPI breaking-change checkers do it:
215
242
 
216
243
  | Class | Changes | Rules |
217
244
  |---|---|---|
218
- | **breaking** (error) | tool removed or renamed, parameter removed, new required parameter, narrower or different type, enum narrowed, a tool becoming less safe (no longer read-only, or additive → destructive) | `diff/tool-removed`, `diff/tool-renamed`, `diff/param-removed`, `diff/param-required`, `diff/param-type`, `diff/enum-narrowed`, `diff/safety-hint` |
219
- | **minor** (info) | tool added, new optional parameter, parameter now optional, type or enum widened | `diff/tool-added`, `diff/param-added`, `diff/param-relaxed`, `diff/type-widened`, `diff/enum-widened` |
220
- | **notice** (info) | optional parameter dropped (extra properties still allowed), description changed (with the text diff: it changes what the agent does), output schema, annotations, other fields, order, a schema restructured but accepting the same input | `diff/param-dropped`, `diff/description`, `diff/schema-other`, `diff/annotations`, `diff/other`, `diff/order`, `diff/schema-equivalent` |
245
+ | **breaking** (error) | tool removed or renamed, parameter removed, new required parameter, narrower or different type, enum narrowed, unlisted properties now rejected (`additionalProperties: false` added), a tool becoming less safe (no longer read-only, or additive → destructive) | `diff/tool-removed`, `diff/tool-renamed`, `diff/param-removed`, `diff/param-required`, `diff/param-type`, `diff/enum-narrowed`, `diff/properties-closed`, `diff/safety-hint` |
246
+ | **minor** (info) | tool added, new optional parameter, parameter now optional, type or enum widened, unlisted properties now accepted (`additionalProperties: false` removed) | `diff/tool-added`, `diff/param-added`, `diff/param-relaxed`, `diff/type-widened`, `diff/enum-widened`, `diff/properties-opened` |
247
+ | **notice** (info) | optional parameter dropped (extra properties still allowed), description changed (with the text diff: it changes what the agent does), output schema, annotations, other fields, order, a schema restructured but accepting the same input, only the declared `$schema` dialect changed | `diff/param-dropped`, `diff/description`, `diff/schema-other`, `diff/annotations`, `diff/other`, `diff/order`, `diff/schema-equivalent`, `diff/schema-dialect` |
221
248
 
222
249
  The same rules apply inside parameters: fields of an object parameter, array items,
223
250
  and the options of an `anyOf`/`oneOf` union (zod's unions, discriminated unions and
@@ -238,11 +265,26 @@ On mongodb-mcp-server 2.1.2 → 3.0.0 this finds `aggregate.pipeline[]…$vector
238
265
  and `.limit` narrowed from `number` to `integer` inside a union, where the whole
239
266
  `pipeline` used to be one "review it".
240
267
 
268
+ Different spellings of the same schema are one schema (`additionalProperties` absent,
269
+ `true` or `{}`; zod 4's safe-integer bounds), and a change every tool shares is said
270
+ once for the menu: a server moving from zod 3 to zod 4 reads as its real changes (a
271
+ field that became required, the patterns zod 4 adds to `email` and `uuid`) plus one
272
+ `diff/properties-opened` line ("27 tools now accept properties they don't list … most
273
+ likely a schema generator upgrade"), not a "review it" per object.
274
+
275
+ The same change in many tools is reported once, with the tools it touches (`pageId`
276
+ is new and required in 25 of 29 tools, on chrome-devtools-mcp 1.8.0); counts are per
277
+ change. When a tool loses a parameter and gains a required one with a close name,
278
+ `diff/param-renamed` (warn, unsure) points out the likely rename; both changes stay
279
+ breaking. Every breaking finding's next step names the version to release when the
280
+ bump is too small.
281
+
241
282
  It also reports the token change per tool and suggests a semver bump. To check the
242
283
  bump, pass the **release** versions: `--release 1.4.0..1.5.0` (npm, a git tag).
243
- `diff/version-bump` then warns when the release's bump is smaller (`0.0.x` promises
244
- nothing and `0.x` may break in a minor), and `diff/version-backwards` when the version
245
- goes down. Calendar versions and prereleases aren't judged. The version in the snapshot is what the
284
+ `diff/version-bump` then warns when the release's bump is smaller. Under 1.0.0 it
285
+ reads versions as npm's caret does (`^0.2.3` accepts any 0.2.x): breaking changes need
286
+ a minor bump, new features a patch, and under 0.1.0 anything goes.
287
+ `diff/version-backwards` warns when the version goes down. Calendar versions and prereleases aren't judged. The version in the snapshot is what the
246
288
  server reports (`serverInfo.version`), which is often not the release: the filesystem
247
289
  server has said `0.2.0` for 19 releases (FINDINGS F5). It's shown, and only checked
248
290
  with `--server-version-is-release`. `diff` also enforces an optional `tokenBudget`
@@ -268,7 +310,9 @@ names, then searches for every operation it finds or sees mentioned, until 20 qu
268
310
  in a row turn up nothing new (at most 200). On Atlassian's hosted server that finds all
269
311
  154 operations behind its 21 tools; on Sentry's, 65 behind 9, the same on every run.
270
312
  Queries are paced and a rate limit is waited out. An operation not found this time is a
271
- notice ("not returned by the same queries"), never "removed". To steer it, set
313
+ notice (`diff/catalog-missing`: "not returned by the same queries"), never "removed";
314
+ one found only now is `diff/catalog-added`, and catalogs read with different queries,
315
+ or partly (a query failed), are `diff/catalog-queries`. To steer it, set
272
316
  `"catalog": { "queries": [...], "maxQueries": 200, "crawl": true }` in
273
317
  `toolmenu.config.json` (FINDINGS F15).
274
318
 
@@ -277,6 +321,17 @@ catalog names, and whose words no query uses, stays unfound. On `@sentry/mcp-ser
277
321
  0.42 (stdio) it finds 63 of the 64 operations in the package's catalog; `whoami` comes
278
322
  back only when asked for by name. Add queries like that to `catalog.queries`.
279
323
 
324
+ **Command routers.** Some servers put their operations behind router tools instead:
325
+ one tool per area that takes a `command` and its `parameters`, and lists its commands
326
+ when called with `learn: true` (Azure's MCP server serves 71 such tools in front of
327
+ ~410 operations). `--catalog` detects them (a `command` parameter, an arguments
328
+ object, a boolean listing flag, and a description that says "router" or "sub
329
+ commands") and calls each once in its listing mode, never with a command, so nothing
330
+ runs. Routers whose command is required aren't called. Detection missed one? Name
331
+ them: `"catalog": { "routers": ["keyvault", "storage"] }`. A router lists everything
332
+ it has, so unlike a search the catalog is complete except for routers that failed
333
+ (`catalog/failed` says which, and why).
334
+
280
335
  ## `history`: a package's releases, researched
281
336
 
282
337
  ```sh
@@ -306,6 +361,9 @@ into description and schema), large enums, and waste you can remove:
306
361
  - **Unused `$defs`**: definitions nothing in the tool refers to. Notion's server 2.5.2
307
362
  ships the same 9 in all 24 tools, 72% of its menu.
308
363
 
364
+ Every tool's estimate is in the menu file, for the ones the summary leaves out:
365
+ `jq -r '.tools[] | "\(.tokens)\t\(.name)"' menu.json | sort -rn`.
366
+
309
367
  ## `snapshot` rules
310
368
 
311
369
  Every rule comes from a failure I actually hit building 115 tools for one MCP
@@ -321,10 +379,11 @@ server, or from the spec itself. If a rule can't point to one, it doesn't ship.
321
379
  | `menu/duplicate-name` | error | Two tools with the same name: only one can be called, and some clients reject the list |
322
380
  | `naming/route` | error | A `routes.yml` expectation broke: a keyword now matches the wrong tool at least as well as the right one |
323
381
  | `description/buried` | warn | Instructions to the agent ("use X instead", "don't retry", "never guess one") past the point where your client cuts descriptions: 2,048 characters in Claude Code, or your `descriptionLimit`. From a real failure: a client cut at 280 characters, and the line that decided routing was at 1,222 |
324
- | `description/cut` | info | Descriptions longer than the client sends, with nothing that reads as an instruction past the cut, as one summary. The model gets a prefix that can read as complete |
325
- | `naming/vague-id` | warn | A parameter called just `id` that doesn't say *which* thing |
382
+ | `description/cut` | info | Descriptions longer than the client sends, with nothing that reads as an instruction past the cut, as one summary that shows where each one is cut. The model gets a prefix that can read as complete |
383
+ | `description/late-instruction` | info | With the default cut: instructions after character 280, which a client that cuts descriptions short would hide, as one summary. Set `descriptionLimit` and `description/buried` checks your real cut instead. (Before 0.13 this was part of `description/cut`; that id's setting still applies.) |
384
+ | `naming/vague-id` | warn | Parameters called just `id` that don't say *which* thing, as one summary with a suggested name |
326
385
  | `ids/authored` | info | An ID the agent must supply that no tool appears to return, so the agent may invent it, one finding per kind of ID (heuristic: IDs from a URL or another tool's text output look the same) |
327
- | `write/unannotated` | warn | A tool named like a write (`delete_`, `send_`…) with no annotations |
386
+ | `write/unannotated` | warn | Tools named like writes (`delete_`, `send_`…) with no annotations, as one summary grouped by the annotation each name suggests |
328
387
  | `naming/shared-word` | info | Nouns that name different things across tools (`list_team_audits` vs `get_audit_trail`), as one summary |
329
388
  | `write/no-dry-run` | info | Destructive tools with no dry-run parameter or preview tool, as one summary |
330
389
  | `spec/discover` | info | The server only speaks the 2025 protocol, so 2026-07-28 rules are skipped |
@@ -402,12 +461,14 @@ toolmenu sees what the server sends. Plenty of agent failures happen elsewhere:
402
461
  --env <K=V> env var for a stdio server, repeatable (it only gets a
403
462
  minimal environment otherwise)
404
463
  --timeout <ms> per-request timeout (default: 30000)
464
+ --protocol <p> auto (default), legacy (2025 handshake only) or modern
465
+ (2026-07-28 only; an error if the server lacks it)
405
466
  --processes <n> server processes or connections to compare (default: 2);
406
467
  1 opens no second one, session's scope check included
407
468
  ```
408
469
 
409
470
  Exit codes: `0` clean · `1` findings at or above `--fail-on` · `2` couldn't connect
410
- or bad usage. `toolmenu --help` lists the `session` and `history` options.
471
+ or bad usage. `toolmenu <command> --help` lists that command's own options.
411
472
 
412
473
  ## Limitations
413
474
 
@@ -426,9 +487,9 @@ or bad usage. `toolmenu --help` lists the `session` and `history` options.
426
487
 
427
488
  ## About
428
489
 
429
- Built by [Niksa](https://niksa.me) while running a 115-tool MCP server and watching
490
+ Built by [Nikša Perović](https://niksa.me) while running a 115-tool MCP server and watching
430
491
  agents trip over its menu. Every rule started as one of those failures, and the
431
- research behind them is in [docs/FINDINGS.md](https://github.com/niksa90/toolmenu/blob/main/docs/FINDINGS.md).
492
+ research behind them is in [docs/FINDINGS.md](https://github.com/niksa90/toolmenu/blob/main/docs/FINDINGS.md). The story of building it, and where it was wrong, is [on niksa.me](https://niksa.me/toolmenu).
432
493
 
433
494
  Found a false positive, or a failure toolmenu should catch?
434
495
  [Open an issue](https://github.com/niksa90/toolmenu/issues) with the menu (a
package/dist/args.d.ts CHANGED
@@ -7,15 +7,41 @@ export type Synthesized = {
7
7
  ok: false;
8
8
  };
9
9
  /**
10
- * A value the schema itself vouches for: `const`, `default`, `examples`, `enum`,
11
- * then by type and format. Never a guess at an ID or anything with a `pattern`:
12
- * a tool that needs one is skipped, not called with junk.
10
+ * Does a value fit the schema's type, enum and const? Only the checks that catch a
11
+ * default the server itself wouldn't accept (Microsoft Learn: `default: null` on a
12
+ * `type: "string"`); patterns and formats are the server's business.
13
+ */
14
+ export declare function fits(value: unknown, schema: JsonSchema | undefined): boolean;
15
+ /**
16
+ * A value the schema itself vouches for: `const`, a `default` or `examples` value
17
+ * that fits the type, `enum`, then by type and format. Never a guess at an ID or
18
+ * anything with a `pattern`, and never an empty list for a list it has to fill: a
19
+ * tool that needs one is skipped, not called with junk.
13
20
  */
14
21
  export declare function synthesize(schema: JsonSchema | undefined, name?: string): Synthesized;
15
- /** Arguments for a tool's required parameters, or the ones it can't fill. */
16
- export declare function synthesizeArgs(schema: JsonSchema | undefined): {
22
+ /**
23
+ * A value the user gave (--value), made to fit the parameter: `5` is the number 5
24
+ * for an integer and the string "5" for a string; JSON that doesn't fit the type
25
+ * falls back to the text as typed.
26
+ */
27
+ export declare function fitValue(given: {
28
+ raw?: string;
29
+ value: unknown;
30
+ }, schema: JsonSchema | undefined): unknown;
31
+ /** Where a tool's argument came from: the user (--value) or the schema. */
32
+ export type ArgSource = 'value' | 'schema';
33
+ /**
34
+ * Arguments for a tool's required parameters, or the ones it can't fill. `given`:
35
+ * values the user supplied for this tool, by parameter name; they come first, and
36
+ * one for an optional parameter is passed too.
37
+ */
38
+ export declare function synthesizeArgs(schema: JsonSchema | undefined, given?: Record<string, {
39
+ raw?: string;
40
+ value: unknown;
41
+ }>): {
17
42
  ok: true;
18
43
  args: Record<string, unknown>;
44
+ sources: Record<string, ArgSource>;
19
45
  } | {
20
46
  ok: false;
21
47
  missing: string[];
package/dist/args.js CHANGED
@@ -3,19 +3,60 @@ const DATE_TIME = '2026-01-01T00:00:00Z';
3
3
  const DATE = '2026-01-01';
4
4
  const SEARCH_WORDS = /^(q|query|search|search_?query|text|keywords?|terms?|prompt|question)$/i;
5
5
  /**
6
- * A value the schema itself vouches for: `const`, `default`, `examples`, `enum`,
7
- * then by type and format. Never a guess at an ID or anything with a `pattern`:
8
- * a tool that needs one is skipped, not called with junk.
6
+ * Does a value fit the schema's type, enum and const? Only the checks that catch a
7
+ * default the server itself wouldn't accept (Microsoft Learn: `default: null` on a
8
+ * `type: "string"`); patterns and formats are the server's business.
9
+ */
10
+ export function fits(value, schema) {
11
+ if (!schema || typeof schema !== 'object')
12
+ return true;
13
+ if ('const' in schema && JSON.stringify(schema.const) !== JSON.stringify(value))
14
+ return false;
15
+ if (schema.enum?.length && !schema.enum.some((e) => JSON.stringify(e) === JSON.stringify(value)))
16
+ return false;
17
+ const branches = (schema.anyOf ?? schema.oneOf);
18
+ if (Array.isArray(branches) && branches.length && !branches.some((b) => fits(value, b)))
19
+ return false;
20
+ if (schema.type === undefined)
21
+ return true;
22
+ const types = Array.isArray(schema.type) ? schema.type : [schema.type];
23
+ return types.some((t) => {
24
+ switch (t) {
25
+ case 'null':
26
+ return value === null;
27
+ case 'string':
28
+ return typeof value === 'string';
29
+ case 'boolean':
30
+ return typeof value === 'boolean';
31
+ case 'integer':
32
+ return Number.isInteger(value);
33
+ case 'number':
34
+ return typeof value === 'number' && Number.isFinite(value);
35
+ case 'array':
36
+ return Array.isArray(value) && (!schema.items || value.every((v) => fits(v, schema.items)));
37
+ case 'object':
38
+ return !!value && typeof value === 'object' && !Array.isArray(value);
39
+ default:
40
+ return true;
41
+ }
42
+ });
43
+ }
44
+ /**
45
+ * A value the schema itself vouches for: `const`, a `default` or `examples` value
46
+ * that fits the type, `enum`, then by type and format. Never a guess at an ID or
47
+ * anything with a `pattern`, and never an empty list for a list it has to fill: a
48
+ * tool that needs one is skipped, not called with junk.
9
49
  */
10
50
  export function synthesize(schema, name = '') {
11
51
  if (!schema || typeof schema !== 'object')
12
52
  return { ok: false };
13
53
  if ('const' in schema)
14
54
  return { ok: true, value: schema.const };
15
- if ('default' in schema)
55
+ if ('default' in schema && fits(schema.default, schema))
16
56
  return { ok: true, value: schema.default };
17
- if (Array.isArray(schema.examples) && schema.examples.length)
18
- return { ok: true, value: schema.examples[0] };
57
+ const example = Array.isArray(schema.examples) ? schema.examples.find((e) => fits(e, schema)) : undefined;
58
+ if (example !== undefined)
59
+ return { ok: true, value: example };
19
60
  if (schema.enum?.length)
20
61
  return { ok: true, value: schema.enum[0] };
21
62
  const branches = (schema.anyOf ?? schema.oneOf);
@@ -53,10 +94,10 @@ export function synthesize(schema, name = '') {
53
94
  return { ok: false };
54
95
  }
55
96
  case 'array': {
97
+ // A required list is there to be filled: `{"urls": []}` is refused by the
98
+ // server (Exa's web_fetch_exa), or does nothing. No item, no call.
56
99
  const item = synthesize(schema.items, name);
57
- if (item.ok)
58
- return { ok: true, value: [item.value] };
59
- return typeof schema.minItems === 'number' && schema.minItems > 0 ? { ok: false } : { ok: true, value: [] };
100
+ return item.ok ? { ok: true, value: [item.value] } : { ok: false };
60
101
  }
61
102
  case 'object': {
62
103
  const args = synthesizeArgs(schema);
@@ -66,16 +107,52 @@ export function synthesize(schema, name = '') {
66
107
  return { ok: false };
67
108
  }
68
109
  }
69
- /** Arguments for a tool's required parameters, or the ones it can't fill. */
70
- export function synthesizeArgs(schema) {
110
+ /**
111
+ * A value the user gave (--value), made to fit the parameter: `5` is the number 5
112
+ * for an integer and the string "5" for a string; JSON that doesn't fit the type
113
+ * falls back to the text as typed.
114
+ */
115
+ export function fitValue(given, schema) {
116
+ if (fits(given.value, schema))
117
+ return given.value;
118
+ if (given.raw !== undefined && fits(given.raw, schema))
119
+ return given.raw;
120
+ // A single value for a list: --value urls=https://… means ["https://…"].
121
+ if (schema?.type === 'array' && fits([given.value], schema))
122
+ return [given.value];
123
+ return given.value;
124
+ }
125
+ /**
126
+ * Arguments for a tool's required parameters, or the ones it can't fill. `given`:
127
+ * values the user supplied for this tool, by parameter name; they come first, and
128
+ * one for an optional parameter is passed too.
129
+ */
130
+ export function synthesizeArgs(schema, given = {}) {
71
131
  const args = {};
132
+ const sources = {};
72
133
  const missing = [];
73
- for (const param of schema?.required ?? []) {
134
+ const required = schema?.required ?? [];
135
+ for (const param of required) {
136
+ if (param in given) {
137
+ args[param] = fitValue(given[param], schema?.properties?.[param]);
138
+ sources[param] = 'value';
139
+ continue;
140
+ }
74
141
  const v = synthesize(schema?.properties?.[param], param);
75
- if (v.ok)
142
+ if (v.ok) {
76
143
  args[param] = v.value;
144
+ sources[param] = 'schema';
145
+ }
77
146
  else
78
147
  missing.push(param);
79
148
  }
80
- return missing.length ? { ok: false, missing } : { ok: true, args };
149
+ if (missing.length)
150
+ return { ok: false, missing };
151
+ for (const [param, g] of Object.entries(given)) {
152
+ if (param in args || !schema?.properties || !(param in schema.properties))
153
+ continue;
154
+ args[param] = fitValue(g, schema.properties[param]);
155
+ sources[param] = 'value';
156
+ }
157
+ return { ok: true, args, sources };
81
158
  }
package/dist/auth.d.ts CHANGED
@@ -6,7 +6,7 @@ import type { OAuthClientInformationContext, OAuthClientMetadata, OAuthClientPro
6
6
  export declare const DEFAULT_PORT = 33418;
7
7
  /** Where logins live: $TOOLMENU_AUTH_DIR, else $XDG_CONFIG_HOME/toolmenu/auth, else ~/.config/toolmenu/auth. */
8
8
  export declare function authDir(): string;
9
- /** Is there a stored login for this server? */
9
+ /** Is there a stored file for this server (a login, or what's left of one)? Why a login is needed reads its contents. */
10
10
  export declare function hasLogin(serverUrl: string): boolean;
11
11
  export interface ProviderOptions {
12
12
  /** A browser login may start (auth login). Otherwise a needed login is an error. */
@@ -28,6 +28,8 @@ export declare class StoredOAuthProvider implements OAuthClientProvider {
28
28
  private readonly options;
29
29
  private data;
30
30
  private expectedState;
31
+ /** invalidateCredentials('all') deleted a file that held tokens, in this run. */
32
+ private removedLogin;
31
33
  constructor(serverUrl: string, options?: ProviderOptions);
32
34
  private stored;
33
35
  private update;
@@ -50,7 +52,7 @@ export declare class StoredOAuthProvider implements OAuthClientProvider {
50
52
  invalidateCredentials(scope: 'all' | 'client' | 'tokens' | 'verifier' | 'discovery'): Promise<void>;
51
53
  }
52
54
  export declare class LoginNeededError extends Error {
53
- constructor(serverUrl: string);
55
+ constructor(serverUrl: string, why?: string);
54
56
  }
55
57
  export interface LoginOptions extends Omit<ProviderOptions, 'interactive' | 'onRedirect'> {
56
58
  /** How to show the authorization URL. Default: print it and try to open a browser. */
package/dist/auth.js CHANGED
@@ -7,6 +7,7 @@ import { homedir } from 'node:os';
7
7
  import { join } from 'node:path';
8
8
  import { Client, StreamableHTTPClientTransport, UnauthorizedError } from '@modelcontextprotocol/client';
9
9
  import { VERSION } from './version.js';
10
+ import { endSession } from './close.js';
10
11
  /**
11
12
  * The redirect port. Fixed, not random: dynamic registration records the exact
12
13
  * redirect URI, so a new port would invalidate the stored client on every login.
@@ -37,7 +38,7 @@ async function save(data) {
37
38
  await writeFile(path, JSON.stringify({ ...data, savedAt: new Date().toISOString() }, null, 2) + '\n', { mode: 0o600 });
38
39
  await chmod(path, 0o600);
39
40
  }
40
- /** Is there a stored login for this server? */
41
+ /** Is there a stored file for this server (a login, or what's left of one)? Why a login is needed reads its contents. */
41
42
  export function hasLogin(serverUrl) {
42
43
  return existsSync(storePath(serverUrl));
43
44
  }
@@ -51,6 +52,8 @@ export class StoredOAuthProvider {
51
52
  options;
52
53
  data;
53
54
  expectedState;
55
+ /** invalidateCredentials('all') deleted a file that held tokens, in this run. */
56
+ removedLogin = false;
54
57
  constructor(serverUrl, options = {}) {
55
58
  this.serverUrl = serverUrl;
56
59
  this.options = options;
@@ -124,11 +127,11 @@ export class StoredOAuthProvider {
124
127
  return (await this.stored()).tokens;
125
128
  }
126
129
  async saveTokens(tokens, ctx) {
127
- await this.update({ tokens, ...(ctx?.issuer ? { issuer: ctx.issuer } : {}) });
130
+ await this.update({ tokens, rejected: undefined, ...(ctx?.issuer ? { issuer: ctx.issuer } : {}) });
128
131
  }
129
132
  async redirectToAuthorization(authorizationUrl) {
130
133
  if (!this.options.interactive) {
131
- throw new LoginNeededError(this.serverUrl);
134
+ throw new LoginNeededError(this.serverUrl, whyLoginNeeded(await this.stored(), this.removedLogin));
132
135
  }
133
136
  await this.options.onRedirect?.(authorizationUrl);
134
137
  }
@@ -138,7 +141,7 @@ export class StoredOAuthProvider {
138
141
  async codeVerifier() {
139
142
  const v = (await this.stored()).codeVerifier;
140
143
  if (!v)
141
- throw new Error('No PKCE code verifier saved for this login.');
144
+ throw new Error(`No PKCE code verifier is saved for ${this.serverUrl}: the stored login changed while this one ran (another login at the same time?).\n→ Next: Run toolmenu auth login ${this.serverUrl} again.`);
142
145
  return v;
143
146
  }
144
147
  async saveDiscoveryState(discoveryState) {
@@ -150,15 +153,22 @@ export class StoredOAuthProvider {
150
153
  async invalidateCredentials(scope) {
151
154
  const s = await this.stored();
152
155
  if (scope === 'all') {
156
+ this.removedLogin ||= Boolean(s.tokens) || Boolean(s.rejected);
153
157
  this.data = { serverUrl: s.serverUrl };
154
158
  await rm(storePath(this.serverUrl), { force: true });
155
159
  return;
156
160
  }
157
161
  const patch = {};
158
- if (scope === 'client')
162
+ if (scope === 'client') {
159
163
  patch.clientInformation = undefined;
160
- if (scope === 'tokens')
164
+ if (s.tokens)
165
+ patch.rejected = 'client';
166
+ }
167
+ if (scope === 'tokens') {
161
168
  patch.tokens = undefined;
169
+ if (s.tokens)
170
+ patch.rejected ??= s.rejected ?? 'refresh';
171
+ }
162
172
  if (scope === 'verifier')
163
173
  patch.codeVerifier = undefined;
164
174
  if (scope === 'discovery')
@@ -166,9 +176,28 @@ export class StoredOAuthProvider {
166
176
  await this.update(patch);
167
177
  }
168
178
  }
179
+ /**
180
+ * Why snapshot or session needs a login, from what's stored: nothing, a registration
181
+ * from a login that never finished, tokens the server refused, or a login the server
182
+ * rejected and toolmenu removed (removedLogin: in this run).
183
+ */
184
+ function whyLoginNeeded(s, removedLogin) {
185
+ if (removedLogin)
186
+ return 'the server rejected the stored login for it, so toolmenu removed it';
187
+ if (s.rejected === 'client')
188
+ return "the stored login no longer works: the server doesn't accept the client toolmenu registered (it may have been deleted), so its tokens were dropped";
189
+ if (s.rejected === 'refresh')
190
+ return 'the stored login no longer works: its token expired or was revoked and the server refused to refresh it';
191
+ if (s.tokens)
192
+ return 'the stored login no longer works: the server refused its token';
193
+ if (s.clientInformation)
194
+ return "there's no stored login for it: an earlier toolmenu auth login registered a client but didn't finish";
195
+ return `there's no stored login for it (stored in ${authDir()})`;
196
+ }
169
197
  export class LoginNeededError extends Error {
170
- constructor(serverUrl) {
171
- super(`${serverUrl} needs an OAuth login, and toolmenu can't open a browser here. Run: toolmenu auth login ${serverUrl}`);
198
+ constructor(serverUrl, why = `there's no stored login for it (stored in ${authDir()})`) {
199
+ super(`${serverUrl} asks for an OAuth login, and ${why}. snapshot and session don't open a browser.\n` +
200
+ `→ Next: Log in once in a terminal with a browser: toolmenu auth login ${serverUrl}. Or pass a token: --header "Authorization: Bearer <token>".`);
172
201
  }
173
202
  }
174
203
  /**
@@ -190,18 +219,22 @@ export async function login(serverUrl, options = {}) {
190
219
  try {
191
220
  await client.connect(transport);
192
221
  // Already logged in with a valid (or refreshed) token.
193
- await client.close();
222
+ await endSession(client, transport);
194
223
  }
195
224
  catch (error) {
196
225
  if (!(error instanceof UnauthorizedError))
197
226
  throw error;
198
227
  const params = await callback.params;
199
- if (params.get('error'))
200
- throw new Error(`The authorization server refused the login: ${params.get('error')}`);
201
- if (!provider.lastState || params.get('state') !== provider.lastState)
202
- throw new Error('The login callback carried the wrong state, so it was ignored. Try again.');
228
+ if (params.get('error')) {
229
+ const described = params.get('error_description');
230
+ throw new Error(`The authorization server refused the login for ${serverUrl}: ${params.get('error')}${described ? ` (“${described}”)` : ''}.\n` +
231
+ `→ Next: Try again. If it repeats, ask for fewer scopes with --scope, or check the account may use this server.`);
232
+ }
233
+ if (!provider.lastState || params.get('state') !== provider.lastState) {
234
+ throw new Error(`The login callback for ${serverUrl} carried the wrong state (not the one this login sent), so it was ignored (an old browser tab, or a second login at once).\n→ Next: Run toolmenu auth login ${serverUrl} again and use the tab it opens.`);
235
+ }
203
236
  await transport.finishAuth(params);
204
- await client.close().catch(() => { });
237
+ await endSession(client, transport);
205
238
  }
206
239
  return await check(serverUrl);
207
240
  }
@@ -210,10 +243,10 @@ export async function login(serverUrl, options = {}) {
210
243
  // register itself. Say what works instead.
211
244
  if (error instanceof Error && /does not support dynamic client registration/i.test(error.message) && !options.clientId) {
212
245
  error.message =
213
- `${serverUrl} doesn't let clients register themselves (no dynamic client registration). Two ways in:\n` +
214
- ` - register an OAuth app with the provider, with the callback URL http://127.0.0.1:${port}/callback, then:\n` +
246
+ `${serverUrl} doesn't let clients register themselves (its authorization server has no dynamic client registration), so toolmenu can't log in without a client of its own.\n` +
247
+ `→ Next: Register an OAuth app with the provider, with the callback URL http://127.0.0.1:${port}/callback, then:\n` +
215
248
  ` toolmenu auth login ${serverUrl} --client-id <id> --client-secret <secret>\n` +
216
- ` - or skip OAuth and pass a token the server accepts: toolmenu snapshot ${serverUrl} --header "Authorization: Bearer <token>"`;
249
+ ` Or skip OAuth and pass a token the server accepts: toolmenu snapshot ${serverUrl} --header "Authorization: Bearer <token>"`;
217
250
  }
218
251
  throw error;
219
252
  }
@@ -224,13 +257,14 @@ export async function login(serverUrl, options = {}) {
224
257
  /** Connect with the stored login and list tools: the login works. */
225
258
  async function check(serverUrl) {
226
259
  const client = new Client({ name: 'toolmenu', version: VERSION });
227
- await client.connect(new StreamableHTTPClientTransport(new URL(serverUrl), { authProvider: new StoredOAuthProvider(serverUrl) }));
260
+ const transport = new StreamableHTTPClientTransport(new URL(serverUrl), { authProvider: new StoredOAuthProvider(serverUrl) });
261
+ await client.connect(transport);
228
262
  try {
229
263
  const { tools } = await client.listTools();
230
264
  return { name: client.getServerVersion()?.name, tools: tools.length };
231
265
  }
232
266
  finally {
233
- await client.close().catch(() => { });
267
+ await endSession(client, transport);
234
268
  }
235
269
  }
236
270
  export async function logout(serverUrl) {
@@ -274,8 +308,12 @@ function listenForCallback(port, timeoutMs) {
274
308
  res.end('<!doctype html><title>toolmenu</title><p>Logged in. You can close this tab and go back to the terminal.</p>');
275
309
  resolveParams(url.searchParams);
276
310
  });
277
- const timer = setTimeout(() => rejectParams(new Error(`No login came back within ${Math.round(timeoutMs / 1000)} s.`)), timeoutMs);
278
- server.on('error', (error) => fail(error.code === 'EADDRINUSE' ? new Error(`Port ${port} is in use: pick another with --port (the login registers that redirect).`) : error));
311
+ const timer = setTimeout(() => rejectParams(new Error(`The browser login didn't come back within ${Math.round(timeoutMs / 1000)} s: nothing reached http://127.0.0.1:${port}/callback.\n` +
312
+ `→ Next: Run toolmenu auth login again and finish it in the browser. The browser has to run on this machine (127.0.0.1); over SSH, log in on your own machine and copy ${authDir()}, or pass a token with --header.`)), timeoutMs);
313
+ server.on('error', (error) => fail(error.code === 'EADDRINUSE'
314
+ ? new Error(`Couldn't wait for the login's redirect: port ${port} on 127.0.0.1 is in use (another toolmenu login, or another program).\n` +
315
+ `→ Next: Close it, or pick another port with --port <n>. The login registers http://127.0.0.1:<n>/callback, so use the same port the next time too.`)
316
+ : error));
279
317
  server.listen(port, '127.0.0.1', () => ready({
280
318
  params,
281
319
  close: () => {