claris-docs

Community-maintained Claude Code plugin for Claris product documentation. Not affiliated with Claris International Inc.

These skills make Claude an authoritative source for Claris/FileMaker questions by fetching official documentation on demand from help.claris.com rather than relying on training data.

Installation

Add the GitHub repo as a marketplace source, then install:

/plugin marketplace add soliantconsulting/soliant-public-claude-plugins
/plugin install claris-docs@soliant-public-claude-plugins

Once installed, skills are available as claris-docs:data-api, claris-docs:filemaker-pro, etc. Skills also trigger automatically when Claude detects a relevant Claris/FileMaker question.

Uninstalling

/plugin uninstall claris-docs@soliant-public-claude-plugins

Updating

Auto-update is off by default for third-party plugins. To enable it, run /plugin, go to the Marketplaces tab, select soliant-public-claude-plugins, and toggle Enable auto-update. When enabled, Claude Code updates the plugin at startup and prompts you to run /reload-plugins.

To update manually:

/plugin marketplace update soliant-public-claude-plugins

Local development

To load the plugin from a local directory without installing it (for development and testing):

claude --plugin-dir /path/to/claris-docs

This loads the plugin for that session only and does not affect your installed version.

Available skills

Skill	Topics
claris-docs:filemaker-pro	Scripting, functions, layouts, relationships, fields, security
claris-docs:filemaker-server	Admin Console, hosting, backups, schedules, monitoring
claris-docs:filemaker-cloud	Cloud setup, administration, Claris ID
claris-docs:filemaker-go	Mobile access, Go development, iOS App SDK
claris-docs:filemaker-webdirect	Browser-based access, deployment
claris-docs:claris-studio	Views, fields, functions, teams
claris-docs:claris-connect	Workflow automation, connectors
claris-docs:claris-customer-console	Accounts, teams, subscriptions
claris-docs:claris-mcp	AI assistant integration via MCP
claris-docs:data-api	REST Data API, authentication, CRUD
claris-docs:admin-api	Server administration REST API
claris-docs:odata	OData protocol, BI tool integration
claris-docs:sql-reference	ExecuteSQL, ODBC/JDBC syntax
claris-docs:security-guide	Encryption, hardening, compliance
claris-docs:app-upgrade-tool	FMUpgradeTool, patch files
claris-docs:data-migration-tool	DMT, data migration
claris-docs:developer-tool	Developer Utilities, binding, runtime

Project setup

For FileMaker projects, copy CLAUDE.md.template into your project root as CLAUDE.md (or append it to an existing one):

cp /path/to/claris-docs/CLAUDE.md.template CLAUDE.md

This gives Claude explicit routing rules so it always consults the right skill before answering from memory — more reliable than auto-trigger alone.

How it works

Skills are navigators, not knowledge stores. They don't embed documentation content.

1. SKILL.md is loaded when a skill is invoked. It tells Claude how to use the reference files.
2. references/ files contain page titles and URLs grouped by topic. Claude reads these to find the right page.
3. Live fetch — Claude fetches the actual documentation page from help.claris.com/markdown/en/{guide}/{page}.md at runtime.

This means answers always reflect the current published docs.

Updating references

Reference files are an index of what's published at help.claris.com. When Claris adds, removes, or renames documentation pages, the references become stale — links may 404 or new content won't be discoverable. Run update.py to resync:

python update.py                          # fetch fresh llms-full.txt from Claris

Or

python update.py snapshots/llms-full.txt  # use a local snapshot

The script regenerates all references/ files deterministically. Git diffs of these files show exactly what changed in Claris's docs.

Design decisions

Progressive disclosure — skills don't dump 1000+ links into context. SKILL.md points to reference files; reference files point to live pages. Each level is loaded only when needed.

Reference file sizing — each file targets ≤80 links as a guideline – strongly encouraged, but not a hard limit. Under 80, a single file is fine (the AI can scan it cheaply). Over 80, split along natural topic boundaries where it makes sense. Over 150, always split. Rationale: each link is ~30-35 tokens; very large unsplit files waste context and risk the AI picking wrong pages. Exception: when no meaningful sub-split exists (e.g. Get functions, which are a single flat category with no sub-grouping), keeping them in one file is preferable to an artificial split. pro-func-get.md (140 links) is a documented exception.

Link-list format — reference files use the same - [Title](URL) format as the source llms-full.txt. Compact, clickable in editors, trivial to emit from update.py.

Skills don't embed content — Claris publishes LLM-friendly Markdown docs. Embedding them would create stale copies. Fetching live means one update.py run keeps the index current while content is always fresh.

404 handling — if a fetch fails, Claude tells the user the page may have moved and suggests running update.py. Skills never silently fail.