Claude plugin

Install

Install the DTPR plugin in Claude Code, Claude Cowork, or Claude.ai, and the capability matrix that determines which skills work fully on each host.

The plugin lives inside the Helpful-Places/dtpr repository under plugin/dtpr/. Installation differs per host.

Any host — `npx skills`

The skills CLI installs DTPR skills into any agent that reads a local skills directory (Claude Code, Cursor, Codex, Cline, Continue, etc.). Skills install as files on disk; the MCP server still has to be wired up separately by your host.

Install every DTPR skill:

npx skills add Helpful-Places/dtpr --all

Install one skill:

npx skills add Helpful-Places/dtpr --skill dtpr-describe-system

List what's available before installing:

npx skills add Helpful-Places/dtpr --list

By default skills writes to the project-local agent directory it detects (e.g. .claude/skills/, .cursor/skills/). Pass --global to install to your home directory instead. See the skills README for symlink-vs-copy and target-agent flags.

npx skills only installs the skill files. The DTPR MCP server at https://api.dtpr.io/mcp is not registered automatically — add it to your host's MCP config manually, or use the Claude Code marketplace install below which bundles .mcp.json.

Claude Code

Add the repository as a marketplace and install:

/plugin marketplace add Helpful-Places/dtpr
/plugin install dtpr

Verify the MCP is live:

/mcp

You should see dtpr as a registered server. /plugin list should show dtpr as installed.

Claude Code is the recommended host for the full authoring loop. All four optional tools (Read, WebFetch, Task, Write) are available, so the research corpus can write new entries via dispatched sub-agents.

Claude Desktop & Claude.ai — bundled zips

For hosts that upload skills one at a time, download a pre-built zip of all seven skills at the current plugin version:

dtpr-skills.zip (v0.3.1) — combined bundle (one .skill file per skill plus an INSTALL.txt).

Unzip and upload each .skill file through your host's skill-upload UI. Individual .skill downloads are also available if you only want one:

The bundle only ships the skill files. Register the DTPR MCP server at https://api.dtpr.io/mcp through your host's MCP configuration — Claude Code's plugin path below does this automatically; Desktop and Claude.ai users add it manually.

A machine-readable manifest with sizes and SHA-256 digests is published alongside the zips at /skills/manifest.json — useful if you script the install or want to verify what you downloaded.

Claude Cowork

Upload the plugin/dtpr/ directory via Cowork's plugin upload path. The bundled .mcp.json registers the remote MCP on install. Tool availability is probed at session start — see the capability matrix below.

Claude.ai

Use the bundled zips above — they're the cleanest path for Claude.ai's skill-upload flow. The MCP tools run over HTTPS and work independently of the skill-upload path. The Task tool is not available on Claude.ai, so the research corpus is effectively read-only from this host — research misses degrade to a logged gap in the output rather than dispatching a sub-agent.

Capability matrix

Each skill probes its needed tools at session start (Phase 0 in dtpr-describe-system; inline degradation in the schema-tier and comprehension skills). The table below is the expected baseline; skills record actuals per session and degrade gracefully when a tool is missing.

Capability	Claude Code	Claude Cowork	Claude.ai
MCP client	✅	✅	✅
`Read` (PDF, local files)	✅	Probe at session start	Unavailable
`WebFetch` (URLs)	✅	Probe at session start	Probe at session start
`Task` (research sub-agents)	✅	Probe at session start	Unavailable
`Write` (corpus entries)	✅	Probe at session start	Read-only
Recommended for	Full authoring loop	Team review; confirm capabilities at first use	Verbal describe + comprehension grading

When a capability is listed as "Probe at session start", the expected behavior is host-dependent and not yet confirmed by a reference session. dtpr-describe-system's Phase 0 trial-call-and-degrade protocol discovers what is actually available and surfaces that in the session narrative.

Verify the install

Any of the seven skills should respond to their trigger phrases after install. A minimal smoke test:

> describe our facial-recognition parking kiosk as a DTPR datachain

If dtpr-describe-system fires and calls list_schema_versions, the plugin is wired up correctly. If no skill fires, try an explicit phrasing (use dtpr-describe-system for …) and check /plugin list on Claude Code or the upload confirmation on Cowork/Claude.ai.

Troubleshoot

Symptom	Check
`/mcp` does not list `dtpr` (Claude Code)	Rerun `/plugin install dtpr`; then `/plugin list` to confirm installation.
MCP tools time out or return 5xx	Hit `https://api.dtpr.io/healthz` — it should return `{"ok":true,…}`. If not, the upstream API is down.
Skills don't trigger on expected prompts	Try an explicit phrasing like "use `dtpr-describe-system` for …". Skill descriptions evolve; see the per-skill pages for current trigger language.
Validation errors reference an unknown element id	Ask the skill to re-list elements (`list_elements`) before constructing the datachain.
Corpus warning "no INDEX.md or malformed"	Normal on a fresh install before any research has been captured. Skills treat this as an empty corpus and continue.
`Task` tool unavailable (Claude.ai)	Research misses log a gap in the output instead of dispatching a sub-agent — expected on hosts without `Task`.

Skills — per-skill reference.
Research corpus — what writes happen, what privacy the _-prefix enforces.
MCP server — the underlying tool registry every skill drives.

Edit this pageorReport an issue

Claude plugin

Seven-skill authoring studio for describing AI systems as DTPR datachains, iterating on the DTPR schema, drafting element icons, translating locales, and grading content for public comprehension.

Skills

The seven peer Agent Skills of the DTPR authoring studio.