Use Hadron slash commands in Claude Code (the hadron plugin)¶

MCP onlyBeginner~5 min

The hadron plugin adds three explicit slash commands to Claude Code that drive Hadron through its MCP tools:

Command	What it does
`/hadron:h-task <name>`	Run a runnable Hadron task by fuzzy name (omit the name to pick from a list).
`/hadron:hadron_search <text>`	Search across the whole Hadron graph — nodes, memories, agents, Apps.
`/hadron:h-open-node <name or urn>`	Find a node and open it in the portal.

Why slash commands and not just plain language? In Claude Code, asking it to "run a task" collides with the built-in Task tool (background agents), so the model tends to launch a subagent instead of calling Hadron. An explicit /hadron:h-task removes the ambiguity — it always calls the right tool.

Before you start¶

Hadron connected via OAuth. The commands call Hadron's MCP tools, so connect Hadron first — follow Add Hadron to Claude Code (OAuth) (one command, one browser sign-in). The plugin does not bundle a connector: you authenticate once at the user level, which also lights up the Claude Chat tab under the same identity.
Claude Code — the CLI or the Code tab of the Claude desktop app. Slash commands are a Claude Code feature; see Surfaces.

Step 1 — Add the marketplace¶

In Claude Code, run:

/plugin marketplace add hadron-memory/hadron-plugins

This is a public repo, fetched anonymously — no GitHub account required.

Step 2 — Install the plugin¶

/plugin install hadron@hadron

Then restart your Claude Code session (or the Code tab) so the new commands register.

Step 3 — Use the commands¶

Run a task¶

/hadron:h-task review-changes

h-task fuzzy-matches the name against the runnable nodes (isRunnable = true) in the memories you can read, and runs the match. Omit the name to get a picker:

/hadron:h-task

If more than one memory or task matches, you'll be asked to choose.

Scope to a memory

Name the memory in the same line — /hadron:h-task review-changes in mmdata — or set an active memory first to skip the memory prompt.

Search the graph¶

/hadron:hadron_search review

Returns one ranked list across organizations, memories, nodes, agents, and Apps.

Open a node in the portal¶

/hadron:h-open-node review-changes

Resolves the name to a node and opens https://hadronmemory.com/app/u/<urn> in your browser. Pass a full hrn:node:… URN to skip the lookup.

Surfaces¶

Slash commands and plugins are a Claude Code feature. Where the plugin works, and where you fall back to plain MCP tools:

Surface	`/hadron:*` commands	Hadron tools
Claude Code CLI	Yes	Yes
Code tab (desktop app)	Yes	Yes
Chat tab (desktop + web)	No — use natural language	Yes, via the OAuth connector
Cowork	Only as a Cowork-format plugin	Yes
Chrome	No	—

On the Chat tab you have the same Hadron tools through the OAuth connector — you just invoke them in plain language instead of with /hadron:*.

Troubleshooting¶

The commands don't appear after installing¶

Restart the Claude Code session (or the Code tab) — plugins load at session start. Confirm with /plugin; the hadron plugin should be listed as installed.

"Run a task" launches a background agent instead¶

That's the Task-tool collision this plugin exists to avoid — type the explicit /hadron:h-task <name> rather than describing it in prose.

A command says "No runnable tasks" or "No memory"¶

The commands only see what your account can read, and h-task only offers nodes flagged isRunnable = true. Confirm you're connected (the verify step in Add Hadron to Claude Code (OAuth)) and that the target memory has runnable nodes.