ICCPlus MCP
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@ICCPlus MCPcreate a new ICC Plus project with a point named 'start'"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
ICCPlus MCP
A schema-aware Model Context Protocol server for creating, inspecting, editing, validating, and packaging ICC Plus projects.
This project was created through vibe coding in collaboration with an AI coding agent.
The server gives an AI agent a complete ICC Plus project model generated and indexed directly from the upstream source:
the project schema and current defaults are generated from ICC Plus source;
every one of the 59 declared model types and 893 unique fields is discoverable;
all 227 authored source/build/config/patch files and 1,411 named functions/methods across the creator and standalone viewer are indexed with exact source, SHA-256 evidence, signatures, model-field usage, and line spans;
all 75 files in the requested deployment repository and 34 files inside its official viewer archives have byte counts and SHA-256 manifests;
high-level tools preserve IDs, ordering, parent links, and reciprocal memberships;
generic RFC 6902 patching keeps new or uncommon upstream fields accessible;
structural and semantic validation catches dangling references before save;
source-backed Custom CSS tools expose official viewer classes, resolve project selectors, and diagnose syntax/cascade risks before save;
ICC Plus v2.10 row-width preservation, build-form visibility/debug titles, split score-recalculation controls, hidden row menus, and addon state CSS classes are modeled and discoverable;
official web and local viewer archives can be built without opening the creator UI.
Compatibility is currently generated from ICC Plus v2.10.1, source commit
b33bfb9b29e0a84a035a56d7e1827e42fe0f7000.
What this is
ICCPlus MCP is an authoring and build adapter around ICC Plus project JSON. It does not reimplement the browser viewer. Selection effects, audio playback, dialogs, rendering, and other interactive behavior still execute in the official viewer. The server understands and validates the configuration that drives those behaviors, can evaluate requirement trees for supplied state, and can package the result into the official viewer templates.
The design deliberately has two layers:
agent
├─ discover: capabilities, schema, resources, prompts
├─ author: create/update/move/duplicate/import entities
├─ complete field access: atomic JSON Patch for every schema field
└─ verify: normalize, validate, evaluate, save, build viewer
│
▼
revisioned in-memory project session
│
▼
ICC Plus JSON / viewer ZIPRelated MCP server: ts-language-mcp
Requirements
Node.js 20 or newer
An ICC Plus project workspace
For viewer builds,
web_viewer.ziporlocal_viewer.zipfrom the ICC Plus deployment repository
Install and build
git clone https://github.com/Kz2g1ew-commits/ICCPlus-MCP.git
cd ICCPlus-MCP
npm ci
npm run buildThe executable is dist/index.js and communicates over stdio. Protocol output
uses stdout; diagnostics use stderr.
Connect an agent
Set ICCPLUS_WORKSPACE to the only directory the server may read or write.
Relative project, asset, template, and output paths are resolved inside this
directory. Existing symlinks are resolved and cannot escape it.
Codex
codex mcp add iccplus \
--env ICCPLUS_WORKSPACE=/absolute/path/to/iccplus-workspace \
-- node /absolute/path/to/ICCPlus-MCP/dist/index.jsVerify with codex mcp list, then restart the client if it was already open.
The equivalent user or trusted-project config.toml entry is:
[mcp_servers.iccplus]
command = "node"
args = ["/absolute/path/to/ICCPlus-MCP/dist/index.js"]
env = { ICCPLUS_WORKSPACE = "/absolute/path/to/iccplus-workspace" }
startup_timeout_sec = 10
tool_timeout_sec = 60
enabled = trueCodex CLI, the IDE extension, and the ChatGPT desktop Codex surface share this
configuration. A project-scoped entry can live in .codex/config.toml and is
loaded only for a trusted project.
Other stdio MCP clients
Use the client's local-server configuration with this command:
{
"command": "node",
"args": ["/absolute/path/to/ICCPlus-MCP/dist/index.js"],
"env": {
"ICCPLUS_WORKSPACE": "/absolute/path/to/iccplus-workspace"
}
}Optional environment:
ICCPLUS_WORKSPACE: filesystem boundary; defaults to the server process CWD.ICCPLUS_MAX_ASSET_BYTES: maximum local asset size; defaults to 26,214,400 bytes (25 MiB).ICCPLUS_DUPLICATED_TEXT_LIMIT_BYTES: maximum size duplicated into the text fallback of a tool result; defaults to 8,192 bytes. Larger exact results remain available instructuredContent. Set-1for legacy unlimited text.
Recommended agent workflow
Call
iccplus_capabilitiesfor the relevant feature family or field.Create or open a project and retain its
project_idandrevision.Create referenced top-level entities first: points, variables, words, groups, designs, and global requirements.
Create rows, then choices, then scores/addons/requirements.
Pass
expected_revisionto mutations. Usedry_runfor broad changes.For advanced styling, discover exact selectors with
iccplus_css_catalog, analyze candidate CSS, then apply it withiccplus_css_set.Call
iccplus_validateand, where useful,iccplus_evaluate_requirements.Save explicitly or build an official viewer archive.
An agent can also request the bundled author-iccplus-project or
audit-iccplus-project prompt.
Example user request:
Create a CYOA with 10 starting credits, three mutually exclusive origin choices, and a weapon row unlocked by selecting an origin. Validate it, save it as
projects/origins.json, then build a web viewer usingtemplates/web_viewer.zip.
Tools
Tool | Purpose |
| Discover 19 feature families, a type/field, an exact function body, a source-file index, or a deployment artifact. |
| List source-backed official viewer classes and exact project-id selectors. |
| Analyze stored or candidate CSS syntax, specificity, target resolution, cascade conflicts, and external assets. |
| Replace, append, prepend, or clear project Custom CSS with revision, dry-run, and validation protection. |
| Read the schema summary, one definition, or the complete generated schema. |
| Start from the exact current upstream defaults. |
| Open project JSON into an isolated session. |
| List sessions, revisions, dirty state, and counts. |
| Inspect validation, size, content counts, and optional JSON. |
| Read exact JSON Pointers, including CSS/HTML/JS and unknown fields, with asset redaction by default. |
| Search all modeled entities by type, ID, or text. |
| Create an entity with defaults, fresh IDs, and parent repair. |
| Deep-merge/unset fields and optionally rewrite ID references. |
| Clone an entity tree with fresh IDs and optional internal remapping. |
| Reorder an entity and repair positional indices. |
| Delete while rejecting newly dangling references by default. |
| Atomically apply RFC 6902 operations to any project field. |
| Migrate legacy shapes and repair IDs, indices, parents, and memberships. |
| Run generated schema plus semantic/reference validation. |
| Evaluate requirements against supplied selections, points, variables, and words. |
| Export one reusable entity subtree. |
| Import a subtree with safe ID regeneration/remapping. |
| Set any string field from a workspace file, URL, or data URL. |
| Atomically save validated JSON inside the workspace. |
| Build official web/local viewer ZIPs and optionally separate images. |
| Undo or redo a session mutation. |
| Close a session, protecting unsaved changes. |
Entity tools cover row, backpack_row, choice, addon,
selectable_addon, score, requirement, point, variable, word,
group, row_design_group, choice_design_group, global_requirement,
sound_effect, and category.
Resources and prompts
URI/name | Contents |
| Official Custom CSS selector catalog with source evidence. |
| Complete generated project JSON Schema. |
| Source-backed feature catalog and coverage counts. |
| SHA-256 manifest for all deployment files and official viewer ZIP entries. |
| UTF-8 normalized metadata for 209 upstream third-party packages. |
| Stored CSS, static analysis, and project-specific selectors. |
| Live revision, validation, and entity counts. |
| Safe construction order and verification workflow. |
| Full project completion and packaging audit. |
Safety and data integrity
Mutations occur on cloned data and commit atomically.
expected_revisionprovides optimistic concurrency control.dry_runreturns the projected validation report without changing a session.no_new_errorsis the default mutation policy;strictrequires a completely valid result;noneis available for staged repairs.JSON Patch blocks prototype-pollution pointer segments.
Save uses a temporary file plus rename.
Existing files are not replaced unless
overwrite=true.Asset values are redacted from ordinary query output and reported by media type and approximate byte size.
Exact path reads avoid returning a complete project, and large tool results are not duplicated into both text and structured output.
Unknown fields survive load, edit, normalization, and save for forward compatibility.
Validation coverage
The validator combines the generated App schema with checks for:
duplicate or missing IDs and positional index mismatches;
addon parent links;
every modeled row, choice, point, group, variable, word, design, sound, and global-requirement reference;
score-to-point and point-activation references;
nested requirements, thresholds, and global-requirement cycles;
group and design-group reciprocal memberships;
incompatible viewer export modes;
Custom CSS syntax, dangerous legacy constructs, unresolved ICC Plus ID selectors, broad scope, external assets, and probable inline-style conflicts;
point integer/float and initialization invariants.
The requirement evaluator implements id (including /ON#N), points,
pointCompare, or, selFromGroups, selFromRows, selFromWhole, gid,
word, negation, operators, and nested prerequisites. It returns an
explainable trace rather than only a boolean.
Advanced Custom CSS
ICC Plus stores CSS in the top-level customCSS field and injects it as
style#customCSS in the official creator/viewer using textContent. The MCP
derives its catalog from the pinned standalone viewer markup, including
row-{id}, row-{id}-bg, row-{id}-header, choice-{id}, choice state
classes, addon, and selectable addon-{id}.
iccplus_css_catalog can return CSS-escaped selectors for every open-project
row, choice, and selectable addon. iccplus_css_analyze parses nested rules,
reports specificity and matched entities, and warns when a declaration is
likely to lose to ICC Plus inline styles. iccplus_css_set persists the result
through the same revisioned transaction and validation policy as other project
mutations.
For v2.10+, the catalog also covers .row-bg-{rowId},
.row-header-{rowId}, .choice, .addon-selectable, and the
.addon-enabled, .addon-disabled, .addon-selected, and
.addon-unselected state classes while retaining older row selector spellings.
This is intentionally static analysis. Computed styles, responsive layout, and interaction-state rendering remain the responsibility of the official viewer; no browser automation runtime or unrelated browsing tool is bundled.
Viewer builds
iccplus_build_viewer accepts an official template ZIP within the workspace.
Web mode writes
project.json.Local mode embeds the project in
js/app.js.Loading title, text, colors, favicon, background, fonts, and project Custom CSS are retained in the packaged project.
Optional image separation extracts data URLs from global styles, rows, backpack rows, design groups, and viewer settings; identical assets are deduplicated.
The template's viewer JavaScript remains the runtime source of truth.
Upstream synchronization
The project model is generated directly from upstream source. Re-run the analyzer against a checkout of ICC-Plus-Svelte:
npm run analyze:upstream -- --source ../ICC-Plus-Svelte --deployment ../ICCPlus
npm test
npm run check
npm run buildThe analyzer uses the TypeScript AST to regenerate:
src/generated/iccplus.schema.jsonsrc/generated/default-project.jsonsrc/generated/source-analysis.jsonsrc/generated/deployment-manifest.jsonsrc/generated/third-party-licenses.jsonanalysis/CODEBASE_INVENTORY.md
Tests fail if a newly declared upstream model type is not assigned to a feature
family. source-analysis.json also records declarations, imports, complete
SHA-256-addressed source files, component labels, every named source
function/method, exact function bodies, and source occurrences. The deployment
manifest covers every file in wahaha303/ICCPlus plus every official viewer
archive entry. Agents can query field:<name>, type:<name>,
function:<name>, or source:<relative-path> through
iccplus_capabilities; deployment:<relative-path> returns deployment
evidence. Function queries support file, offset, and limit so duplicate
local names remain manageable; include_source=false returns compact metadata.
Development
npm test
npm run test:stdio
npm run verify:upstream -- --source ../ICC-Plus-Svelte --deployment ../ICCPlus
npm run check
npm run build
npm run inspect
npm pack --dry-runThe test suite covers generated-model coverage, model graph mutation, normalization, reference rewrites, all requirement families, RFC 6902 atomicity, viewer packaging, and an in-memory MCP client/server exchange.
For the full semantic feature map, see
analysis/FEATURE_ANALYSIS.md. The generated
field/function/component evidence is in
analysis/CODEBASE_INVENTORY.md.
The requirement-by-requirement verification record is
analysis/COMPLETION_AUDIT.md.
Attribution
ICC Plus is developed by
wahaha303. This MCP server is an independent
integration and does not replace or modify the official creator/viewer.
Both projects are distributed under the MIT License; see LICENSE.
Upstream and vendored dependency notices are preserved in
THIRD_PARTY_NOTICES.md and the
iccplus://licenses resource.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- AlicenseNot gradedqualityBmaintenanceEnables local-first book writing with AI assistance, allowing users to scaffold, create, search, validate, and enrich a book repository through natural language using OpenCode or Claude.MIT
- AlicenseAqualityCmaintenanceEnables AI coding agents to interact with TypeScript projects through compiler-level code intelligence, providing tools for navigation, type information, diagnostics, refactoring, and semantic search.293393Apache 2.0
- AlicenseAqualityAmaintenanceMCP server for AI agents to compile, playtest, and exhaustively explore ink interactive fiction stories, catching compile errors, runtime errors, unreachable content, and verifying story structure.5592MIT
- AlicenseAqualityBmaintenanceEnables AI tools to collaboratively write novels by managing chapters, characters, and story state through commands like validate, context, draft, review, and approve.8MIT
Related MCP Connectors
Give your AI agent a persistent map of your project's structure, dependencies, and bugs.
Build, validate, and deploy multi-agent AI solutions from any AI environment.
Cross-agent artifact workspace with provenance across Claude Code, Codex, Cursor, LangGraph.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/Kz2g1ew-commits/ICCPlus-MCP'
If you have feedback or need assistance with the MCP directory API, please join our Discord server