@apicity/cli

One command per upstream endpoint, for every @apicity/* provider package — plus the agent skill that teaches a coding agent to find and call them.

The endpoint list is the monorepo's scripts/endpoint-docs.tsv (bundled as dist/endpoint-docs.tsv for installed users), so the command surface stays in lockstep with the providers: no curated subset, no new abstractions.

Renamed from @apicity/mcp-server. MCP support was removed in September 2026; MCP.md is the migration note.

Install

npm install -g @apicity/cli
# or run it without installing
npx -y @apicity/cli@latest commands --provider openai

Use @latest with npx; bare npx -y @apicity/cli can reuse an older cached package that does not know newer flags. Node.js 20 or newer.

Calling an endpoint

apicity <provider> <dotPath> [--method M] [--<param> <value>…] [--data <json>] [global flags]

<dotPath> is the endpoint's path in the provider tree, exactly as scripts/endpoint-docs.tsv and the provider READMEs spell it:

apicity xai v1.tokenizeText --data '{"model":"grok-3","text":"Hello, world!"}'
apicity openai v1.models --method GET
apicity binance api.v3.time

Discovery comes first and costs nothing — no credential, no network, no provider package loaded:

apicity providers                     # every provider, its env vars, whether set
apicity commands --provider openligadb # that provider's endpoints
apicity describe xai v1.tokenizeText  # one endpoint: schema, example, call shape
apicity xai v1.tokenizeText --help    # the same description, from the call form

A dotPath that carries more than one HTTP method exits ambiguous (8) until --method picks one. apicity <provider> on its own is shorthand for apicity commands --provider <provider>.

Parameters: three kinds, one grammar

Every value is passed as --<name> <value> (or --<name>=<value>), and what the CLI does with it comes from the generated call-shape table, never from the URL:

• Path parameters are passed to the endpoint function as arguments. apicity describe lists them as pathParams.

  apicity openai v1.files.content --method GET --id file-abc123

• Request fields are merged into the request object, so a flag and --data are interchangeable. apicity describe lists them as requestParams.

  apicity s3 objects.get --bucket my-bucket --key photo.png
  apicity s3 objects.get --data '{"bucket":"my-bucket","key":"photo.png"}'

• Provider-supplied placeholders are not flags at all. Telegram's {token} and thesportsdb's {apiKey} are substituted by the factory from the credential, so naming one is a usage error rather than a token silently forwarded or ignored.

Both spellings of a name work: the URL's placeholder (--vector_store_id) and the property the endpoint declares (--vectorStoreId).

The request body

--data '<json>', --data-file <path>, or --data - to read stdin — that last one is the only stdin the CLI ever reads, and nothing prompts. Passing both --data and --data-file is a usage error, as is a body that is not valid JSON.

Global flags

Every flag below works on the endpoint form. An endpoint's own parameter flags come from its call shape, so a provider that ships a --json parameter tomorrow cannot collide with this table.

The built-ins read their own smaller sets, because each has argv the shared table has no business interpreting: commands, describe and providers take --json and --quiet plus --provider and --method, and --env-file, --op-vault and --op-token, which decide what they count as configured; skill, setup and doctor take --json and --remove, and setup 1password also --no-verify. Passing one of the flags below to a built-in that does not know it is a usage error (--verbose needs a value, exit 1) rather than a silent no-op, and doctor names an argument it ignored on stderr.

| Flag | Description | | ------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | --json | Force the machine envelope, even at a terminal. With --quiet, force compact JSON. | | --quiet | Print the data alone, without the envelope. | | --verbose | One line per HTTP call on stderr: method, URL, status, elapsed. Never a header or a body. | | --method <M> | Pick one method when a dotPath answers several. | | --data <json> | The request body; - reads stdin. | | --data-file <path> | The request body, read from a file; - reads stdin. | | --otp <token> | A paid endpoint's single-use approval, needed only when the operator armed the gate; refused with no secret (Paid endpoints). | | --output-dir <path> | Where binary results and downloaded media land. Also APICITY_OUTPUT_DIR. | | --env-file <path> | Load provider settings from a dotenv file. Set env vars win; op:// references resolve (see 1Password). | | --op-vault <vault> | The vault convention: fill missing credentials from op://<vault>/<ENV_VAR>/password. Also APICITY_OP_VAULT. | | --op-token <token> | 1Password service-account token: a literal, env:VAR, $VAR, or a variable name. It also authenticates op:// references. --op-service-token is an alias. Also APICITY_OP_SERVICE_TOKEN. | | --paygate-secret-file <path> | Arms the pay gate: the shared HMAC secret that verifies paid-endpoint OTPs. Unset, the gate is off. Also APICITY_PAYGATE_SECRET_FILE. | | --base-url <url> | Override the provider's base URL (see the note below). | | --timeout <ms> | Override the provider's request timeout, in milliseconds. | | --help, -h | On the endpoint form, describe the endpoint; otherwise print usage. | | --version | Print the package version. |

--base-url and --timeout reach the addressed factory, and support is not uniform. They are merged into the factory's options as baseURL (baseUrl for zaicoding, the one factory that spells it that way) and timeout in milliseconds. A factory that reads neither ignores them silently rather than failing; a few — alibaba, for one — compose a second upstream base from the origin of what you pass rather than its whole path; and an endpoint whose upstream lives on its own host (xai's management API, fireworks' inference base) keeps that host whatever --base-url says. --verbose prints the URL each call actually requested, which is how to check rather than guess.

Built-in commands

| Command | What it does | | -------------------------------- | --------------------------------------------------------------------------------------------------------------------- | | apicity commands | Every endpoint, or one provider's with --provider. --json for a catalog. | | apicity describe <p> <dotPath> | One endpoint: method, URL, docs URL, path parameters, request parameters, whether it is paid, and the request schema. | | apicity providers | Every provider, the env vars it reads (names only) and whether it is configured. | | apicity skill | Print the agent skill this package ships, byte for byte. | | apicity skill install | Install that skill into ~/.agents/skills/apicity. --remove undoes it. | | apicity setup claude | Install the skill and the Claude Code plugin. | | apicity setup codex | Install the skill Codex reads. | | apicity setup agents | Install the skill; connect an agent when exactly one is unambiguous. | | apicity setup 1password | Save --op-vault and --op-token to the env file and check them once; --no-verify, --remove. | | apicity setup | setup agents. --remove removes the plugin and everything this CLI wrote. | | apicity doctor | Twelve rows about this host's install, always exit 0. | | apicity help [topic] | Usage, or one of the topics output, exit-codes, environment, agents. | | apicity version | The package version. |

Environment variables

| Variable | Effect | | ----------------------------- | -------------------------------------------------------------------------- | | APICITY_ENV_FILE | The dotenv file to load when --env-file is absent. | | APICITY_OP_VAULT | The 1Password vault, as --op-vault; apicity setup 1password saves it. | | APICITY_OP_SERVICE_TOKEN | The 1Password token, as --op-token; apicity setup 1password saves it. | | APICITY_PAYGATE_SECRET_FILE | The pay-gate secret file that arms the gate, as --paygate-secret-file. | | APICITY_OUTPUT_DIR | Where results land, as --output-dir. | | APICITY_SETUP_AGENT | claude, codex, all or none — what apicity setup agents connects. | | CLAUDE_PROJECT_DIR | Read by Claude Code sessions; the output directory when nothing else says. | | Provider variables | One or more per provider; see Credentials. |

With no --env-file and no APICITY_ENV_FILE, a call loads ~/.config/apicity/.env when that file exists. For each variable of the addressed provider — that provider alone, never all 28 — the first source that has one wins:

1. a value already set in the environment;
2. a literal in the env file;
3. an op:// reference, the environment's before the env file's, resolved through op (see 1Password);
4. the vault convention, op://<vault>/<VAR>/password, when a vault and a token are both configured.

Exit codes

apicity help exit-codes prints this table. Branch on the code field of the error envelope rather than on the number: several codes share a status and the numbers may gain siblings.

code                    exit
0                       0
usage, skill_unmanaged  1
not_found               2
auth                    3
forbidden, paygate      4
rate_limit              5
network                 6
api, setup_incomplete   7
ambiguous               8

| code | Raised when | | ------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | usage | An unknown flag, a missing required --<param>, invalid --data, an unreadable --data-file, or a refused flag combination. | | skill_unmanaged | A skill directory exists that this CLI did not write, so it will not be overwritten. | | not_found | An unknown provider, command or dotPath; a --method naming a method the dotPath does not answer (exit 2; hint lists the methods it does); or upstream answered HTTP 404. | | auth | The addressed provider has no credential configured, a 1Password read failed, or upstream answered HTTP 401. | | forbidden | Upstream answered HTTP 403. | | paygate | The operator's armed gate refused a missing, malformed, expired, mismatched or replayed OTP, or an --otp reached a host with no secret. Its code is in hint. | | rate_limit | Upstream answered HTTP 429. | | network | The request never landed: DNS failure, refused connection, abort or timeout. | | api | Any other upstream non-2xx status or provider exception, and local operational failures such as an unwritable output directory. | | setup_incomplete | An agent that was asked for explicitly is not installed, or the packaged skill is missing. | | ambiguous | The dotPath answers several methods and --method was not given. |

Credentials

Every provider reads its credential from the environment. The CLI prints variable names and whether they are set — never a value, on any command, in any mode.

| Provider | Env var | | ------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | openai | OPENAI_API_KEY | | xai | XAI_API_KEY | | anthropic | ANTHROPIC_API_KEY | | fireworks | FIREWORKS_API_KEY | | fal | FAL_API_KEY | | google | GOOGLE_API_KEY | | googleflow | GOOGLE_FLOW_API_KEY | | dolthub | DOLTHUB_API_KEY | | dropbox | DROPBOX_OAUTH_TOKEN | | simplefunctions | SIMPLEFUNCTIONS_API_KEY (optional — keyless endpoints work without it) | | kie | KIE_API_KEY | | kimicoding | KIMI_CODING_API_KEY | | alibaba | DASHSCOPE_API_KEY | | zaicoding | ZAI_CODING_PLAN_API_KEY | | binance | (none — public API) | | openligadb | (none — public API) | | openf1 | (none — public API) | | elevenlabs | ELEVENLABS_API_KEY | | s3 | S3_ACCESS_KEY_ID + S3_SECRET_ACCESS_KEY | | b2 | B2_ACCESS_KEY_ID + B2_SECRET_ACCESS_KEY + B2_REGION | | x | X_ACCESS_TOKEN | | meta | IG_ACCESS_TOKEN | | polymarket | POLYMARKET_CLOB_API_KEY + POLYMARKET_CLOB_API_SECRET + POLYMARKET_CLOB_API_PASSPHRASE + POLYMARKET_ADDRESS + POLYMARKET_PRIVATE_KEY + POLYMARKET_FUNDER_ADDRESS (public market data needs none) | | free-media-upload | (none — public API) | | youtube | YOUTUBE_ACCESS_TOKEN (optional — keyless endpoints work without it) | | telegram | TELEGRAM_BOT_KEY | | quo | QUO_API_KEY | | thesportsdb | THESPORTSDB_API_KEY (optional — keyless endpoints work without it) |

apicity providers prints this table for the machine you are on, with a configured column. POLYMARKET_SIGNATURE_TYPE is public configuration rather than a secret: set it to the account's verified value (0, 1, 2 or 3) alongside the credential bundle, as a literal or an explicit op:// reference. Polymarket refuses to load a credential bundle without it.

1Password

An env file can hold op://<vault>/<item>/<field> references instead of literals. That is the format op run --env-file reads, and the one this repository's own .env uses, where an item's title need not match its variable:

FIREWORKS_API_KEY=op://Apicity/FIREWORKS_AI_API_KEY/password

A call resolves the references of the provider it addresses, and only those, in one op inject however many there are. op authenticates with the configured service-account token when there is one (--op-token, APICITY_OP_SERVICE_TOKEN or the env file), and otherwise with its own sign-in: OP_SERVICE_ACCOUNT_TOKEN, the desktop app, or op signin. A reference names its own vault, so it needs no --op-vault. One that cannot be resolved is exit 3 auth, naming the variable and the reference, never a value. In addition to the credential variables listed above, explicit references resolve for these existing factory settings:

| Provider | Additional reference variables | | ------------ | ------------------------------ | | s3 | S3_REGION, S3_ENDPOINT | | b2 | B2_ENDPOINT | | polymarket | POLYMARKET_SIGNATURE_TYPE | | All others | None |

Settings and credentials share the same reference batch and precedence rules. These settings are not added to the credential requirements, the offline configured check, or automatic vault-convention lookup. Omitted settings retain their existing defaults and factory validation. B2_REGION remains part of B2's required credential-variable set. FIREWORKS_ACCOUNT_ID and XAI_MANAGEMENT_API_KEY are not CLI factory inputs and are not resolved; this does not introduce account-ID defaults or management-key wiring.

The vault convention fills what is still missing from op://<vault>/<VAR>/password, once a vault and a token are both configured. An item the vault lacks is exit 3 auth, naming the variable and the reference it expected, except for the four providers whose credential is optional: polymarket, simplefunctions, thesportsdb and youtube take the items the vault has and skip the rest, so their keyless endpoints work with a vault that holds none of them. A vault that holds part of polymarket's bundle resolves that part, and polymarket then applies its own rule: without POLYMARKET_SIGNATURE_TYPE it refuses the bundle with exit 7 api. Save the pair once, and every later call on this host uses it:

apicity setup 1password --op-vault Apicity --op-token env:OP_SERVICE_ACCOUNT_TOKEN

That writes APICITY_OP_VAULT and APICITY_OP_SERVICE_TOKEN to the env file a call loads, replacing those two lines in place and leaving every other line as it was. The token is saved exactly as given, so the env:VAR form keeps the secret itself off disk: prefer it to a literal. A directory the command creates is 0700 and a file 0600; an existing file keeps its mode. It asks nothing: the vault and token come from the flags, else from those two variables in the environment, never from the file. It then checks the pair once with op item list, with the token in op's environment rather than on its command line, and answers exit 7 when op is missing, 3 when 1Password refuses the pair and 6 on a timeout; the lines stay written either way. --no-verify skips that check, and apicity setup 1password --remove deletes the two lines and nothing else.

In apicity providers, commands and describe, configured means the CLI knows where each credential comes from: a variable in the environment, a literal or op:// reference in the env file, or the vault convention when a vault and a token are both configured. Those commands never run op, so whether the vault really holds an item shows at call time and in apicity doctor.

Output

One rule decides everything the CLI prints: --json, or a stdout that is not a terminal, selects the machine envelopes; a terminal without --json selects the pretty form. Both carry the same data.

// success, on stdout
{ "ok": true, "data": { … }, "summary": "…", "meta": { … } }

// failure, on stderr — one line, and nothing on stdout
{ "ok": false, "error": "…", "code": "…", "hint": "…", "meta": { … } }

The discovery commands answer the same envelope. commands, describe and providers — and the apicity <provider> and --help forms that reuse them — put their result under data like every other command (skill install, setup, doctor, and any endpoint call), so a jq filter reads it there:

apicity commands --provider openligadb --json | jq '.data | length'
apicity describe kie api.v1.jobs.createTask --json | jq '.data.paid'

At a terminal without --json, commands and providers print an aligned table and describe a text block instead. A failure follows the one rule above on every path, discovery included: the error envelope whenever stdout is a pipe or --json is passed, and Error: plus hint: lines at a terminal without it.

Success goes to stdout and failure to stderr, never both for one invocation. --quiet prints the data alone — compactly with --json, so apicity … --json --quiet is one line a shell can hand to another program.

Media and binary results are persisted, not printed. A binary response (openai v1.audio.speech, say) is written to the output directory and the data becomes { savedTo, bytes }. A JSON response is walked for https:// strings under the media keys url, downloadUrl, download_url, audio_url, audioUrl, video_url, videoUrl, image_url, imageUrl, fileUrl and file_url; each is downloaded and a sibling <key>_savedTo added next to it. A download that fails is inlined as <key>_savedTo: "error: …" and never fails the call. The directory is --output-dir, else APICITY_OUTPUT_DIR, else CLAUDE_PROJECT_DIR, else the working directory.

Streaming endpoints are buffered into an array rather than streamed: the CLI prints one document.

Paid endpoints

A few endpoints cost money on the provider's side: the rows apicity describe marks paid (paid: yes, or "paid": true with --json), which is also the paid column of apicity commands. Whether or not this host gates them, that flag is the cue to say a call bills before making it.

The pay gate in front of them is opt-in. With no --paygate-secret-file and no APICITY_PAYGATE_SECRET_FILE, the gate is off and a paid call goes upstream directly, with no --otp. An --otp given on such a host exits 4 paygate, with paygate-not-configured repeated in hint: it is refused rather than silently dropped, because whoever passes one believes a gate exists. A secret file that reads as empty is exit 1 usage (<path> is empty), raised before any provider is built, never a disarmed gate.

Naming a secret file arms the gate, and its guarantees are unchanged. The CLI is then the code client: it holds the shared HMAC secret to verify an OTP, every paid call needs one, and each OTP is single-use and bound to the exact request. The CLI never mints one, and grep -rn mintOtp packages/cli/src finds nothing. On a host whose operator armed the gate, a human mints the OTP out-of-band from the same secret, and the caller passes it unchanged:

apicity-paygate otp mint \
  --secret-file ./paygate.secret \
  --dot-path api.v1.jobs.createTask \
  --payload-file request.json \
  --ttl 10m

apicity kie api.v1.jobs.createTask --data-file request.json --otp <token>

On an armed host a missing or bad OTP fails closed with exit 4 and the gate's own code (otp-missing, otp-expired, otp-replayed, …) repeated in hint. The agent driving the CLI never sees the secret, so it cannot self-approve. See @apicity/cost for the full spec.

The skill and the Claude Code plugin

apicity skill prints the agent skill this package ships — the guide that teaches an agent the three discovery steps and the call. apicity skill install writes it to ~/.agents/skills/apicity, the location Codex and other agents read directly, and links ~/.claude/skills/apicity at it for Claude Code.

Every directory this CLI writes carries .managed-by-apicity and an .installed-version stamp. Shape is not ownership: a skill directory without that marker was written by someone else, so the CLI refuses to overwrite or remove it (skill_unmanaged, exit 1) and --remove reports it under kept.

apicity setup claude   # skill + marketplace + plugin install
apicity setup codex    # skill; nothing is ever written under ~/.codex
apicity setup agents   # skill, plus whichever agent is unambiguous here
apicity setup --remove # uninstall the plugin, remove only what this CLI wrote

apicity setup 1password is the fourth form, and it touches only the env file: see 1Password. Its --remove deletes the two lines it wrote, --no-verify skips its op item list check, and no other form's --remove ever opens the env file.

setup claude installs the skill first and unconditionally — the half that works on every host — then runs claude plugin marketplace add, marketplace update and plugin install, and decides the outcome by re-reading installed_plugins.json rather than by trusting an exit status. It is idempotent. APICITY_SETUP_AGENT (claude, codex, all, none) overrides detection for setup agents; a typo warns and falls back to detection, because that command's whole contract is that it finishes without asking anybody anything.

apicity doctor reports twelve fixed rows — CLI and Node versions, env file, 1Password CLI and vault, pay-gate secret, output directory, providers configured, skill install, skill freshness, Claude plugin and Codex — each ok, warning or error with a hint. A check that does not apply to this host says so at ok rather than disappearing, so .data has the same twelve indices everywhere. It always exits 0: it is a report, not a gate. A skill older than the installed CLI is a warning that names apicity skill install; nothing re-syncs itself behind your back. The pay-gate secret row is ok when no secret file is named by flag, the environment or the env file, because the gate is then off; it is error for a file that is unreadable or empty, and it never prints the file's contents.

The 1Password row finds the vault and the token as a call does, flag first, then the environment, then the env file, and names where the token came from: env:VAR, $VAR or a set variable name is shown, and a literal is only ever called one. A token alone is ok, since op:// references resolve with it. A vault alone is an error, because every call then exits usage. With both, doctor runs op item list on the vault once and warns, by name, about each provider variable that nothing else supplies and the vault has no item for. The Providers row counts providers the way apicity providers does, offline.

What we took from hey-cli

This CLI is modelled on basecamp/hey-cli, and the debt is worth naming precisely. Adopted: one binary whose subcommands document themselves through --help; the success envelope {ok, data, summary, meta} and the error envelope {ok, error, code, hint, meta} with the same field names and semantics; --quiet for the data alone; stable exit codes 0–8 printed by apicity help exit-codes as hey help exit-codes prints hey's; a commands --json catalog, here derived from endpoint-docs.tsv; the help topics, as output, exit-codes, environment and agents (hey's linked-accounts has no apicity equivalent); doctor with its {name, status, message, hint} rows; skill and skill install with an ownership marker and a version stamp, spelled .managed-by-apicity; setup claude, setup codex and setup agents with --remove and an APICITY_SETUP_AGENT override where hey reads HEY_SETUP_AGENT; the .claude-plugin/ layout — plugin.json, the skills link and the SessionStart liveness hook — plus a marketplace manifest so the repository installs; and the README-and-AGENTS.md guidance for agents, here as the root README's "CLI and coding agents", this file, and the apicity CLI sections in .claude/CLAUDE.md and AGENTS.md. Adapted: hey's styled terminal output and JSON-when-piped rule, which here prints pretty data at a terminal and the envelope everywhere else, with --json forcing it; hey's .surface snapshot and check-surface-compat, replaced by a catalog-parity test against endpoint-docs.tsv plus a pinned list of built-in commands; and hey's automatic once-per-release skill re-sync, replaced by doctor reporting a stale install for apicity skill install to refresh. hey's API-COVERAGE.md and STYLE.md needed no equivalent: scripts/endpoint-docs.tsv already is the inventory, and this repository's conventions live in .claude/CLAUDE.md.

Not adopted, and why. --jq: --quiet piped into real jq does the job, so the built-in filter can wait. --ids-only, --count, --markdown, --html, --styled and --stats: results here are arbitrary provider JSON with no common list or id shape to shape. The first-run wizard, the interactive sign-in offer and HEY_NONINTERACTIVE: apicity authenticates with API keys the operator configures, there is nothing to sign into, nothing prompts, and bare apicity prints help. Browser OAuth, the system keyring and the auth command family: environment variables, a .env file and 1Password are the credential story. The TUI, watch, upgrade, the installer scripts and the packaging matrix (Homebrew, mise, Nix, deb, Scoop, Omarchy), the .size-budget and the separate skills-sync repository: npm distributes this package, and the skill ships inside it. hey's mcp subcommand: the operator removed MCP support entirely on 2026-09-17, so there is no server to expose.

Switching from @apicity/mcp-server

@apicity/mcp-server was renamed @apicity/cli, and MCP support ended with @apicity/mcp-server 0.11.2: the CLI has no server to register.

1. Install the CLI. npm install -g @apicity/cli (or run it through npx -y @apicity/cli@latest).

2. Check the host. apicity doctor reports the CLI and Node versions, the env file, 1Password, the pay-gate secret, the output directory, how many providers are configured, and the skill and plugin install — one row each, --json for a machine.

3. Connect the agents. apicity setup claude installs the skill and the Claude Code plugin; apicity setup codex installs the skill Codex reads. apicity setup agents does whichever is unambiguous on this host.

4. Remove the old MCP registration. claude mcp remove apicity, codex mcp remove apicity, or delete the mcpServers.apicity entry from the client's config file. Agents reach the endpoints through the CLI and the skill from here on.

5. Retire a launcher script. A launcher that execs the old server bin from node_modules/.bin has nothing left to launch: remove it together with its @apicity/mcp-server dependency rather than repointing it. This repository's own city launcher is one such script.

6. Tell the agents that cannot install plugins. Add a short section to the project's AGENTS.md (and CLAUDE.md or .claude/CLAUDE.md, for Claude Code):

   apicity is driven through the apicity CLI on PATH, never an MCP server. Run apicity --help for the command set and apicity skill for the agent guide (apicity commands, apicity describe, then the call). Credentials are configured by the operator; never request, copy, or read them.

7. After the release, the operator deprecates the old package — a manual, post-publish step no part of the build performs:

   npm deprecate @apicity/mcp-server "Replaced by @apicity/cli; MCP support was removed, use the apicity CLI and agent skill"

Programmatic use

import { loadCatalog } from "@apicity/cli";

const endpoints = await loadCatalog();
console.log(endpoints.filter((e) => e.provider === "openai").length);

runMain(), zodToJsonSchema(), loadEnvFile() and the 1Password helpers are exported too. For diagnostics and bulk checks, buildRegistry() resolves the tsv in one pass against every provider it can instantiate, skipping any whose required credential is unset; loadCatalog() above is the cheap path.

Part of the apicity monorepo.

License

MIT — see LICENSE.