> ## Documentation Index
> Fetch the complete documentation index at: https://arka-agent-mintlify-f21cea86.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Use every Arka skill through MCP

> Discover and invoke dispatch-backed Arka capabilities from MCP clients.

This page covers dispatch-backed skills and opt-in personal tools. Routing table and session flow: [AI agent guide](/guides/ai-agents).

Arka exposes three generic MCP entry points:

* `arka_capabilities` lists the current MCP tools and dispatch-backed skills.
* `arka_skill` invokes any dispatch command with a skill name and arguments.
* `arka_route` accepts natural language and routes it through Arka's complete
  symbolic/fallback router.

Use `arka_route` as the umbrella MCP skill when a client is unsure which
specific Arka tool applies. Send the complete user request as `prompt`; do not
collapse it into a guessed skill name like `design_from_screenshot` or
`web_answer`.

This means newly added skills—including data collection, media transforms,
agent races, reusable blocks, ultra-fast development, OCR, RAG, and future
skills—do not need a separate MCP wrapper. Execution still passes through
Arka's normal mode, security, and write-action gates.

## Local-file MCP tools

`arka_ocr` and `arka_rag` are dedicated MCP tools for OCR and document RAG.
They only work when the agent can read paths on the local machine (for example
a Cursor workspace file). Cloud agents without mounted files cannot use them.

Call `arka_capabilities` to see the `local_file_tools` section, which lists
these tools and the shared notice agents should follow.

### Incremental verification (all agents)

Do **not** wait for an entire long log or batch job before checking results.

1. Run the smallest useful demo first (one file, one page, or one sample).
2. Inspect that result. Confirm success or failure from the output.
3. If the first demo succeeds, run a second increment.
4. Only then report the workflow as **verified**.

`arka_capabilities` exposes this under `agent_execution_rules.incremental_verify`.

| Tool | Local path required | Typical use |
| - | - | - |
| `arka_ocr` | Always (`path`) | Extract text from images; OCR scanned PDFs |
| `arka_rag` | For `ingest`, `codebase_ingest`, `batch_ingest` | Ingest PDFs/docs and ask questions over them |

Example MCP calls:

```json theme={null}
{ "path": "/Users/me/workspace/scan.png", "action": "extract" }
```

```json theme={null}
{ "action": "ingest", "path": "/Users/me/Documents/report.pdf" }
```

```json theme={null}
{ "action": "ask", "document": "report.pdf", "question": "Summarize key findings" }
```

**Defaults (2026):** `PDF_RAG_PGPT=auto` skips PrivateGPT auto-start when TurboQuant
already indexed the file. Scanned PDFs are OCR'd locally before indexing. RAG `ask`
falls back to Ollama, then extractive context, if cloud LLM routes fail.

MCP also applies a safe default filter for personal desktop/device skills.
Spotify playback/search, Brave/browser-opening helpers, media playback, and
`daily_brief` are hidden or blocked unless the MCP server is started with
`ARKA_MCP_ENABLE_PERSONAL_SKILLS=1` or a narrow
`ARKA_MCP_ENABLED_TOOLS` / `ARKA_MCP_ENABLED_SKILLS` override.

## Routing cheat sheet

| Situation | Tool |
| - | - |
| Unsure which Arka tool fits | `arka_route` with full user `prompt` |
| Know the skill name | `arka_skill` with `skill` + optional `args` |
| Need live tool list + rules | `arka_capabilities` |
| Repo layout / symbols | `arka_repo_map` |
| Fuzzy-find a project folder | `arka_tech_stack` (`action: search`) |
| OCR or document RAG | `arka_ocr`, `arka_rag` (local paths) |
| Patch files | `arka_edit_guard` → `arka_apply_patch` |

Full decision table and examples: [AI agent guide](/guides/ai-agents#routing-decision-table).


## Related topics

- [AI agent guide — use Arka over MCP](/guides/ai-agents.md)
- [Arka — AI terminal agent documentation](/index.md)
- [Arka CLI command and flag reference](/guides/cli.md)
- [How Arka compares to other agents](/concepts/comparison.md)
- [Symbolic and LLM routing pipeline](/concepts/routing.md)


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.