Repository navigation
Conversation
…ult to a file (#2431) Writes the single-result path's output to a file instead of stdout. json is the whole result pretty-printed; raw is the text of the text-bearing blocks, or the decoded bytes of a single binary block. Under --format json stdout carries an { output } envelope in place of { result }. Flag combinations that would be silently inert are rejected up front. The TUI half is split to #2571. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
`--quiet` drops the CLI's non-essential output so a scripted caller gets the result payload on stdout and, on failure, only the error envelope on stderr: - a stdio server's own stderr (piped and drained instead of inherited) - the non-strict schema-portability hint on tools/list - the --verify one-line summary (a failing run's envelope carries it) - "Authorization complete." / "Authorization complete. Retrying…" - the --relogin revocation-failure warning Kept on purpose: the --strict report (explicitly requested; the detail behind exit 6), and everything a human must act on for interactive OAuth (authorization URL, step-up [y/N], "open it by hand"). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…2434) `mcp-inspector --cli --completion <bash|zsh|fish>` prints a completion script and exits without connecting. The flag list, which flags take a value, and which take a path are read from the CLI's own commander program, so the scripts cannot drift from --help. --method completes the same ONE_SHOT_METHODS list parseArgs validates, plus servers/list and servers/show; --transport, --log-level, --format and --protocol-era complete their accepted values. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
Since the OAuth store was re-keyed per authorization server (#1625), acquired tokens live at servers[url].byIssuer[activeIssuer].tokens, but the CLI's stored-auth lookups still read the server-level tokens field. --use-stored-auth, --list-stored-auth and --wait-for-auth therefore never saw a token the store actually held. Add resolveStoredCredentials, which resolves a stored server's tokens and client information the way the shared store answers a ctx-less read: the activeIssuer slot first, then the legacy top-level fallback, never an arbitrary issuer. Tokens and client come from the same source, with a preregistered (static) client winning as it does in the auth provider. Every stored-auth lookup goes through it, and the refresh write-back persists rotated tokens to the slot they were read from. Closes #2517 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
A usage error made Commander write its own error line during parse(), before the envelope, so --quiet stderr was two lines. The error is still thrown and reaches the envelope with the same message. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…2435) resolveSecretStore() announces a keychain fallback or a store caveat (e.g. memory) with console.warn on first use, so an HTTP/SSE --quiet run on a box without a keychain still printed it. Core caches the resolution, so the CLI now settles it once at parse time with console.warn muted; no core change, and the selected store is unaffected. Also reword the README so it no longer says --quiet empties stderr, since the table below lists what it deliberately keeps. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
Bash splits `--method=tools/` on '=' (COMP_WORDBREAKS), so the value completes against the option two words back; zsh keeps it as one word, so the option prefix moves into IPREFIX via compset. Fish already handles the form natively; all three are covered by shell tests. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…2433) The CLI could only read the server catalog. Add --method servers/add, servers/edit and servers/remove, which write the same catalog file the web and TUI clients use by driving the web backend's own /api/servers routes in-process (createRemoteApp + app.request), so validation, the secret-store split and the atomic write are shared rather than re-implemented. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
Core reports advisories with console.warn from many places the CLI reaches (cleanRoots, OAuth endpoint overrides, oauth-persist, file-lock, secret-store selection), so muting only the secret-store resolution left the rest leaking. parseArgs now swaps console.warn for a no-op when -q is in argv, and runCli restores it in finally so a module caller (the launcher, the test runner) never inherits the mute. Replaces the narrower quiet-secret-store helper. Everything --quiet keeps is written to the streams directly, never through console.warn. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
… not found A server name that matches an inherited Object.prototype member (`constructor`, `toString`, ...) resolved to the inherited function on a bare `map[name]` read, so the CLI failed with an unrelated "Cannot use 'in' operator" error instead of "Server 'constructor' not found". Both by-name lookups on the --server path now go through the existing `getOwnEntry` helper (an `Object.hasOwn` read): - `selectServerEntry` in core/mcp/node/servers.ts, which is the lookup the CLI's `--server` actually reaches (and `servers list <name>`); - `loadServerFromConfig` in core/mcp/node/config.ts, the lookup named in the issue (web `--config` + `--server` single mode). Closes #2537 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
A rename moves the entry's secrets to the new name, but a session store in the CLI process is empty, so the move would drop secrets another process holds. Also reword the README's servers/remove sentence to name the flags it actually rejects. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
Commander accepts -qe KEY=V, which enabled quiet without the pre-parse handling (Commander's writeErr and the console.warn mute). Read clusters the way Commander does: q is the flag, and e takes a value, so anything after it in the token is that value. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
#2431) Buffer.from(x, 'base64') silently drops invalid characters, so a malformed payload was written as unrelated bytes. Decode through core's base64ToBytes (atob, which throws) and report output_not_raw instead. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…can (#2435) A -q that is another option's value (--client-secret -q) was taken for the flag. Commander's error output is now buffered and written from exitOverride unless opts().quiet is already set, and the console.warn mute moves to just after parse(). Combined -qe still counts because Commander parses it natively; argvRequestsQuiet is gone. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
The shared transport builder drops -e/--cwd for an SSE/HTTP target, so servers/add (or an edit with a new URL) reported success without storing them. A whitespace-only --rename was likewise silently ignored. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…found An explicitly blank --config fell through to the default catalog, and a non-object catalog entry passed the existence precheck only to hit the route's idempotent DELETE and report success without writing. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…lags (#2420) Web exposes the per-server skills catalog budget (skillCatalogMaxSkills / skillCatalogMaxBytes) in Server Settings; the CLI and TUI already honored the values from a catalog/config file entry but had no flag for them. Add both flags to the CLI and TUI, validated as positive integers by a shared core coerce (skillCatalogLimitParser), and overlay them onto the resolved server settings in loadServerEntries the way --protocol-era is, so verifySkills reads them through getServerSettings(). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…lookup' into v2/chore/2568-cli-rollup
…suer' into v2/chore/2568-cli-rollup
…o v2/chore/2568-cli-rollup
…n' into v2/chore/2568-cli-rollup
…into v2/chore/2568-cli-rollup Signed-off-by: cliffhall <cliff@futurescale.com> # Conflicts: # clients/cli/README.md
…to v2/chore/2568-cli-rollup Signed-off-by: cliffhall <cliff@futurescale.com> # Conflicts: # clients/cli/README.md # clients/cli/__tests__/README.md # clients/cli/src/cli.ts # clients/cli/src/handlers/method-types.ts
…t-cli-tui' into v2/chore/2568-cli-rollup
…et flag error (#2420) Copilot review: `1e3` and `9007199254740993` are positive integers that the parser rejects by design, so 'Expected a positive integer' did not tell the user how to correct them. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…lls pane (#2420) Copilot review: SkillsTab verifies one skill per gesture, and the catalog budget is a run-level bound, so the TUI flags (like the file values) have no observable effect there today. Say so in the README and --help rather than claiming the pane reads within the budget; making the pane verify the whole catalog is tracked in #2590. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…ader (#2420) Copilot review: tui.tsx is outside the TUI coverage include and tui-servers.test.ts calls the loader directly, so a misspelt option or a dropped forwarding field would have passed. Mock the loader to capture the options runTui builds and assert both flags arrive parsed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…t-cli-tui' into v2/chore/2568-cli-rollup
Copilot review: the budget is read only by verifySkills, so accepting the flags on a run without --verify would make a job look bounded when nothing is bounded. Reject them there, as --require-digests already is. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Signed-off-by: cliffhall <cliff@futurescale.com>
…t-cli-tui' into v2/chore/2568-cli-rollup
Member
Author
|
Closing: this rollup did its job. It merged all seven #2568 sub-issue PRs onto |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Refs #2568
Rollup — not for merge. This branch merges all seven #2568 sub-issue PRs onto
v2/mainto show they combine cleanly and pass CI together. Each change lands through its own PR; close this one once they have merged.--server constructor→ not found)byIssuerslot-q/--quiet--completion <bash|zsh|fish>servers/add,servers/edit,servers/remove--output <path>/--output-format raw|jsonMerge conflicts resolved here
All were additive. Both sides were kept in each:
clients/cli/README.md: the "Editing the server catalog" (CLI has no write path for the server catalog (add/remove/edit servers) #2433) and "Quiet output" (CLI has no --quiet/--silent flag to suppress non-essential output #2435) sections, and the--quiet/--output/--output-formatrows in the options table.clients/cli/__tests__/README.md: theservers-write.test.tsandoutput-file.test.tsrows.clients/cli/src/cli.ts: the--quietand--output/--output-formatoption definitions, the option-type fields, and theMethodArgspassthrough.clients/cli/src/handlers/method-types.ts: thequietandoutput/outputFormatfields.#2537 and #2420 both touch
core/mcp/node/{config,servers}.ts. They merged without conflicts.Local verification on the combined tree
clients/cli:npm run check(format, lint, typecheck) passes.clients/tui:npm run checkpasses.clients/web:tsc -bpasses.core/mcp/nodeunit tests: 150 of 150 pass.clients/clitest:coverageon the first six branches: 566 of 566 pass, with thresholds met.CI on this PR is the check for everything combined.
🤖 Generated with Claude Code