{
  "markdown": "# LlamaParse Agent Plugins\n\nFour document-focused plugins from LlamaIndex, packaged as a marketplace for both Claude and Codex.\nParse documents locally with LiteParse, or connect to the LlamaParse Platform for cloud parsing and\nagentic retrieval over Index v2 knowledge bases.\n\n| Plugin | Type | What it does | Needs an account? |\n| --- | --- | --- | --- |\n| `liteparse` | Skill | Local OCR parsing for PDFs, Office docs, and images | No — fully local |\n| `llamaparse` | Skill | Cloud parsing via the LlamaParse API | Yes — API key |\n| `llamaparse-mcp` | MCP + skills | Parse/split/classify plus Index v2 retrieval | Yes — OAuth |\n| `llamaparse-mcp-eu` | MCP + skills | The same, pointed at the Europe region | Yes — OAuth |\n\n**Which MCP plugin?** Region follows your LlamaCloud account, not a per-session choice. Accounts on\n`cloud.llamaindex.ai` need `llamaparse-mcp`; accounts on `cloud.eu.llamaindex.ai` need\n`llamaparse-mcp-eu`. A token issued in one region is rejected by the other.\n\n## Install\n\n<details>\n<summary><b>Claude</b></summary>\n\nAdd the marketplace:\n\n```\n/plugin marketplace add run-llama/llamaparse-agent-plugins\n```\n\nThen install whichever plugins you want:\n\n```\n/plugin install liteparse@llamaparse-marketplace\n/plugin install llamaparse@llamaparse-marketplace\n/plugin install llamaparse-mcp@llamaparse-marketplace\n/plugin install llamaparse-mcp-eu@llamaparse-marketplace\n```\n\nSkills are namespaced by plugin, so invoke them as `/<plugin>:<skill>`:\n\n```\n/liteparse:liteparse\n/llamaparse:llamaparse\n/llamaparse-mcp:llamaparse-mcp   # parse / split / classify\n/llamaparse-mcp:llamacloud-index # Index v2 agentic retrieval\n```\n\nThe `llamaparse-mcp-eu` plugin exposes the same skills under its own namespace\n(`/llamaparse-mcp-eu:llamaparse-mcp`, and so on).\n\nIf the install summary says `Run /reload-plugins to activate`, run it.\n\n</details>\n\n<details>\n<summary><b>Codex</b></summary>\n\nAdd the marketplace:\n\n```bash\ncodex plugin marketplace add run-llama/llamaparse-agent-plugins\n```\n\nThen start or reload Codex, open plugin management, find the marketplace `llamaparse-marketplace`\n(display name: `LlamaParse Codex MarketPlace`), and install any of `liteparse`, `llamaparse`,\n`llamaparse-mcp`, or `llamaparse-mcp-eu`.\n\nConfirm availability in a Codex thread — skills appear as `liteparse:liteparse`,\n`llamaparse-mcp:llamaparse-mcp`, `llamaparse-mcp:llamacloud-index`, and so on.\n\nThe MCP plugins declare `authentication: ON_USE` in the Codex marketplace, so Codex prompts on the\nfirst tool call rather than at install time.\n\n</details>\n\n## Authentication\n\n- `liteparse` — none. Runs fully locally.\n- `llamaparse` — needs a [LlamaParse API key](https://cloud.llamaindex.ai/signup) in\n  `LLAMA_CLOUD_API_KEY`.\n- `llamaparse-mcp` / `llamaparse-mcp-eu` — OAuth, no API key. The flow runs in your client on first\n  tool use.\n\n## Verify it works\n\n<details>\n<summary><b>LlamaParse MCP — document processing</b></summary>\n\n1. Request an upload URL with `getUploadUrl`.\n2. Upload a file (for example `test.txt`) to the returned pre-signed endpoint.\n3. Call `parseFile` with the returned `fileId`.\n\nExpected result: parsed text content is returned.\n\n</details>\n\n<details>\n<summary><b>LlamaParse MCP — Index v2 retrieval</b></summary>\n\n1. Call `listIndexes` to confirm at least one index is visible.\n2. Call `retrieveFromIndex` with a natural-language question against one of the returned indexes.\n\nExpected result: relevant passages from your indexed documents are returned.\n\nAn empty `listIndexes` result still means auth succeeded — you just have no indexes yet.\n\n</details>\n\n## Privacy Policy & Terms\n\n`liteparse` runs entirely locally and transmits no data. The `llamaparse`, `llamaparse-mcp`, and\n`llamaparse-mcp-eu` plugins send the documents you ask them to process to the LlamaParse Platform\nfor parsing, indexing, and retrieval.\n\n- [Privacy Notice](https://www.llamaindex.ai/legal/privacy-notice)\n- [Terms of Service](https://www.llamaindex.ai/legal/terms-of-service)\n- Support: support@llamaindex.ai\n\n## Repository layout\n\n```\n.claude-plugin/marketplace.json    marketplace definition consumed by Claude\n.agents/plugins/marketplace.json   marketplace definition consumed by Codex\nplugins/<name>/\n  .claude-plugin/plugin.json       Claude manifest\n  .codex-plugin/plugin.json        Codex manifest\n  .mcp.json                        MCP server config (MCP plugins only)\n  skills/<skill>/SKILL.md          skills, one directory each\n  assets/                          icon and logo\n```\n\nEach plugin carries a manifest per agent. The two must agree on name and version — the validator\nenforces this.\n\n## Development\n\n### Validate\n\nCI runs this on every pull request, and it is worth running before you push:\n\n```bash\npython3 scripts/validate-marketplace.py\n```\n\nIt checks that both marketplace manifests list the same plugins, that every registered plugin\nexists on disk (and vice versa), that each `plugin.json` name matches its directory, that the\nClaude and Codex versions of a plugin agree, that `.mcp.json` and the inline `mcpServers` block\ndeclare the same server, that no two plugins claim the same MCP server key, and that every skill\ndirectory has a `SKILL.md`.\n\nFor Claude specifically, also run the first-party validator — the plugin directory's review\npipeline runs the same check:\n\n```bash\nclaude plugin validate .\nclaude plugin validate ./plugins/<name>\n```\n\n### Bump version\n\n```bash\npython3 scripts/bump-version.py <claude|codex> <plugin> --bump <patch|minor|major>\n# if you have uv\n./scripts/bump-version.py <claude|codex> <plugin> --bump <patch|minor|major>\n```\n\nBump both agents so the versions stay in sync:\n\n```bash\n./scripts/bump-version.py codex liteparse --bump minor\n./scripts/bump-version.py claude liteparse --bump minor\n```\n\n### Keeping skills in sync\n\nMost skills here are **copies** with an upstream elsewhere. There is no automation, so they drift\nsilently. Check them before a release.\n\n| Skill | Upstream | Relationship |\n| --- | --- | --- |\n| `liteparse/skills/liteparse` | [`llamaparse-agent-skills`](https://github.com/run-llama/llamaparse-agent-skills/blob/main/skills/liteparse/SKILL.md) | Verbatim copy — sync as-is |\n| `llamaparse/skills/llamaparse` | [`llamaparse-agent-skills`](https://github.com/run-llama/llamaparse-agent-skills/blob/main/skills/llamaparse/SKILL.md) | Verbatim copy — sync as-is |\n| `llamaparse-mcp*/skills/llamaparse-mcp` | [`mcp-llamaindex-ai`](https://github.com/run-llama/mcp-llamaindex-ai/blob/main/skills/llamaparse-mcp/SKILL.md) | Copy with intentional deltas — see below |\n| `llamaparse-mcp*/skills/llamacloud-index` | [`llamaparse-agent-skills`](https://github.com/run-llama/llamaparse-agent-skills/blob/main/skills/llamacloud-index/SKILL.md) | Deliberate rewrite — do **not** sync |\n\n`llamacloud-index` shares a name with the upstream skill but is a different document: upstream\ndrives the REST API with `curl` and an API key, while the version here drives the MCP tools over\nOAuth. They are not interchangeable — treat them as separate skills that happen to share a title.\n\nTwo deltas in `llamaparse-mcp` are intentional and should survive a sync: the `Index v2 Retrieval`\nsection (which points at the sibling `llamacloud-index` skill, something the upstream server repo\nhas no notion of), and the cross-region `401` advice (upstream tells the user to repoint their\nclient, which a plugin user cannot do because the URL is fixed in the manifest — here the\ninstruction is to install the sibling regional plugin).\n\nThe EU plugin's skills are copies of the NA plugin's with the hostname and region swapped. Change\none, change the other.\n",
  "bytes": 7620,
  "sha": "2ff118341e7e5268df171391cd2bf0556bcc79f7c7c2fbb0ce7e19d2d0b0fce4",
  "repo_slug": "run-llama/llamaparse-agent-plugins",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/plg_run_llama_llamaparse_agent_plugins_llama_02b00428/readme"
}