npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

@inventarch/mcp-door

v3.8.1

Published

[![npm](https://img.shields.io/npm/v/@inventarch/mcp-door?label=npm)](https://www.npmjs.com/package/@inventarch/mcp-door) [![Public quality](https://github.com/inventarch/ia/actions/workflows/platform-quality.yml/badge.svg?branch=main)](https://github.com

Readme

@inventarch/mcp-door

npm Public quality License Node

ia-mcp-door is a local MCP server that gives an agent host access to IA workspaces over stdio. It serves twenty tools: eleven read-only Door operations from @inventarch/runtime, the same reads the ia CLI performs, ia_vocabulary, five scoped ia_kb_* knowledge readers, and three method tools, ia_session, ia_catalogue and ia_method, which carry a method kit's procedure to any host.

ia host claude and ia host codex register it for you. Any other MCP client can launch it directly.

⚡ Quick start

With Claude Code or Codex. In an initialized project, preview the registration, then apply it:

ia host claude
ia host claude --apply --yes

Claude Code gets an ia-workspace server in .mcp.json; ia host codex --apply --yes adds [mcp_servers."ia-workspace"] to .codex/config.toml. Both entries start the copy of this server bundled in the host payload that ia host installs under ~/.ia/hosts/ (or $IA_HOME/hosts/), with your Node binary and the project's absolute path as --root. The files are machine-local; do not commit them. To commit a door entry instead, run ia host claude --commit (or codex or copilot): it adds an ia server that runs npx -y @inventarch/mcp-door@<version> from the project folder, with no local path, beside an instruction block that names ia_session. Claude Code asks for approval before it starts a project MCP server.

With any MCP client. Install the package and point the server at a workspace:

npm install --global @inventarch/mcp-door
ia-mcp-door --root path/to/your-project

Without --root it serves the current directory. --kits <folder> (or the IA_KITS_DIR environment variable) names a folder of method kits, such as the ia plugin's config/kits; the method tools also read the kits installed under <IA home>/kits/<folder>/<version>/. At a workspace whose .ia/distributions.lock.json installs kits from an npm package of kits (the package entry ia install --package writes), they read those kits from the workspace store, where the engine installed them as ordinary releases, and never from another copy. A fresh clone holds no store yet: ia_session says to run ia restore, which refetches them by digest. It speaks MCP 2025-11-25 as UTF-8 JSON-RPC, one message per line on stdin and stdout. A client sends initialize and notifications/initialized before tools/list or tools/call. In the project from the root quick start, this session reads the authored @decision:

printf '%s\n' \
  '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"demo","version":"1"}}}' \
  '{"jsonrpc":"2.0","method":"notifications/initialized"}' \
  '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"ia_read","arguments":{"locator":"work-system/definition/decision/api-response-format"}}}' \
  | ia-mcp-door --root .

The second response line carries the result as structuredContent (shortened here):

{"jsonrpc":"2.0","id":2,"result":{"content":[…],"structuredContent":{"ok":true,"result":{…,"kind":"record","digest":"dae12679…","body":"Choose a response format for the public API.","certified":false}},"isError":false}}

🧭 What it does

| Tool | Returns | | --- | --- | | ia_scope | A scope token, or one narrowed by phase, identities or root. Pass it as within to later calls. | | ia_context | Cited procedure cells and applicable governance for a declared phase and primitive. format: "compact" summarizes omissions. | | ia_select | Exactly one of the identities you supply; a tie or a missing coordinate refuses. | | ia_get | One admitted record. | | ia_records | The admitted record snapshot. It can be large; ia_search, ia_get and ia_traverse read less. | | ia_resolve | The identity a typed reference resolves to; an unresolved reference says why. | | ia_search | Ranked matches in decoded record text, with statistics computed inside the scope. | | ia_traverse | Relationships followed from start records, with typed filters and optional condition gating. | | ia_read | The body behind a locator (<identity>, <identity>#<phase>/<Primitive>, <identity>#<REQ-ID> or <path>:<line>), with the SHA-256 of its UTF-8 bytes and certified: false. includeRuntime: true reads a record at runtime placement. | | ia_next | One plan's delivery view, as ia next prints it: milestones and tasks in prerequisite order, each task's verdict, and next, the command for the first task with no declared blocker. | | ia_position | {body, digest, hostNote} for a scope key: seat (an identity or {path}), shape, phase, depth (0 to 2), budget (0 to 64) and word, plus optional primitive. Omitted, primitive is derived from shape; explicit primitive also supplies the default phase. No key part means the repository's @workspace at context, orient, depth 0, budget 0. | | ia_vocabulary | The public words and their schemas, from the catalogue shipped with the server. It reads no workspace and takes no scope token. | | ia_session | {sessionId, root, workspace, kits, lanes, pin?, restore?, briefing}: the kits found (bundled, installed or home), restore with the kits the lock installs that the store lacks and a briefing line that says to run ia restore, the lanes work, knowledge and delivery, the task pin last written for the workspace, and a briefing of at most 2,500 characters. It appends a line to <IA home>/sessions-v1/<sessionId>.jsonl. claim: {path, mode} is recorded as advisory; nothing is fenced. | | ia_catalogue | {lane, items} for lane (work, knowledge or delivery), optionally narrowed by word and phase. Each item has a key such as work.brainstorm, its display name (ia-work:brainstorm, or ia-work for work.*), its kit, version, capability and profile, a one-line description and its state, bound or installed. | | ia_method | The method packet for one key, exactly as the ia plugin's resolve.mjs prints it, with the same digest; a refusal is {error: {code, message, candidates?, key?}}, such as IA-CORE-NO-METHOD. With task, it writes the task pin under <IA home>/work-v1/<project>/<task>/, <project> being the workspace folder's name as a slug. subject is kept only as its SHA-256. |

The first eight tools were introduced in machine-protocol version 1 and ia_read, ia_next and ia_position in version 2. The server uses the latest protocol, version 3, whose ia_position schema additionally accepts primitive. A runtime Door explicitly serving version 2 still refuses that parameter. The protocol's report operation has no tool; run ia report from the CLI.

Every result carries its JSON as structuredContent and as text. A refusal sets isError: true. Door tools return ok: false with a top-level code; knowledge tools return {version: 1, ok: false, error: {code, message}}. Refusals from the three version 2 tools can also carry next, the one command to run:

  • ia_read: IA-RUNTIME-READ-UNADMITTED for a locator no admitted record in the scope answers. Through a narrowed token it names nothing outside the scope; through the whole workspace it says what the sources hold there, such as a record whose admission was refused.
  • ia_next: IA-RUNTIME-NEXT-SEAT, -NO-PLAN, -AMBIGUOUS or -CYCLE, with the plans or the cycle concerned.
  • ia_position: IA-RUNTIME-REQUEST-INVALID for an invalid key or an unknown parameter, such as free text.

The ia_position body is a pure function of the key and the admitted records. It names no root, token or time, so two agents asking the same key of the same records read the same body and digest. hostNote sits beside it and says how fresh the last ia capture is.

⚖️ What it does not do

  • It never writes a workspace. The read tools are annotated readOnlyHint: true, and the workspace files are unchanged after a session. The method tools write only outside it (each *-v1 home folder named here is the authoritative one unless its plain folder holds an authoritative marker; see the runtime SPEC): the session log and the task pin under the IA home, retained method bytes under its method-closures-v1, and a kit's restored read copy under its kit-roots-v1. Authoring, validation and capture stay with the CLI (ia preview, ia author, ia validate, ia capture).
  • It checks mandates only for the method tools. A method tool acts at the server's --root or at a root the IA home's workspace-locations.json lists, and only when that workspace's governing mandates admit Observation for its default actor. The read tools check no @mandate; the CLI's mandate-checked commands, the edit guard and ia-runner do. Mandates are cooperative restrictions, not authentication or a sandbox.
  • It grants nothing. ia_select and ia_context choose and deliver records; they execute nothing. A read certifies nothing: the digest identifies bytes, not whether they are true.
  • Scope tokens are process-local and can expire. A token from ia_scope lasts at most as long as the server process. Once 4,096 newer scopes are issued or read before it is read again, it refuses with IA-DB-STALE (IA-KB-SCOPE-UNAVAILABLE from an ia_kb tool) and reads nothing. Call ia_scope for a new one, with the same narrowing. Calls that omit within read the initial boundary, which never expires. Narrowing a token limits what later calls read through it.
  • Bounded transport. Stdio only, with no network listener. A line over 4 MiB or invalid UTF-8 ends the session, and tools/list issues no pagination cursor.

📚 Learn more

Five additional read tools—ia_kb_check, ia_kb_search, ia_kb_read, ia_kb_resolve, and ia_kb_traverse—read selected native records and their bound documents through the knowledge foundation. They accept an optional within issued by this server. The original Door validates membership, and a matching graph revision is required before any body read. A stale, foreign or unsupported narrowed scope refuses; there is no workspace fallback. The complete knowledge-tool result envelope, including both text and structured content, is bounded to 48 KiB; an oversized request id instead receives a JSON-RPC error. Local citations resolve only while current captured bytes still match. These tools create no index or .kb directory.