Documentation
Everything you need to connect Nimrod and put it to work. Support: admin@orchis.ai.
1 · Getting connected
claude.ai and Claude Desktop (recommended)
- Open Settings → Connectors → Add custom connector
- Paste
https://nimrod.orchis.ai/mcp - Sign in with your email — a six-digit code arrives, and a free 50-credit trial account is created automatically. No card, no API key handling.
Claude Code
Get an API key from the portal (issued automatically on your first visit), then:
claude mcp add nimrod --transport http https://nimrod.orchis.ai/mcp \ --header "Authorization: Bearer nmk_your_key"
Nimrod Desktop — the full toolkit (Windows · macOS · Linux)
A locally-installed MCP server: search is metered through your key, but webpage extraction and research synthesis run on your machine (always free and private — extracted pages never transit our servers). Six research skills, a deep-research agent, and two search-coach hooks auto-install into your Claude setup on first run.
- Claude Desktop: download nimrod.mcpb, double-click, paste your API key when prompted
- Claude Code: download for
Windows ·
macOS (universal) ·
Linux,
run
nimrod-desktop setup(stores and validates your key — validation is free), thenclaude mcp add nimrod -- path/to/nimrod-desktop. On macOS/Linuxchmod +xthe download first; macOS also needs a one-timexattr -d com.apple.quarantine nimrod-desktop-macos(the binary is not yet Apple-signed). On Windows, SmartScreen may warn about a new publisher the first time — choose More info → Run anyway.
Desktop commands:
setup ·
status ·
diagnose (free connectivity test) ·
init/uninstall (install lifecycle) ·
skill <name> ·
hook <name>.
Installed skills: deep-research, fact-check, find-docs, news-monitor,
competitive-analysis, cite-sources — invoke them in Claude with
/deep-research <topic> etc.
Hooks: search-craft
(fires before google_search) and
research-scope (before research_topic)
inject query-crafting and scoping guidance automatically in Claude Code, so every
metered call is a considered one. They're registered in
~/.claude/settings.json at install, merged alongside any
hooks you already have, and removed exactly by
nimrod-desktop uninstall.
Other MCP clients (VS Code, Gemini CLI, LM Studio…)
Nimrod speaks standard MCP, so most MCP-capable clients work. Two patterns:
- Remote (Streamable HTTP): point the client at
https://nimrod.orchis.ai/mcpwith headerAuthorization: Bearer nmk_your_key— this is how VS Code (Copilot agent mode), Gemini CLI, and Cursor connect. - Local (stdio): register the
nimrod-desktopbinary as a stdio MCP server — works in LM Studio, Gemini CLI, VS Code, and anything else that launches local servers. Extraction stays free and local.
The four research tools work identically in every client. Skills, the deep-research agent, and search-coach hooks are Claude Code features — other clients get the tools without the coaching layer.
2 · Tool reference
| tool | what it does | key parameters | cost |
|---|---|---|---|
| google_search | Organic Google results with quality scores, authority ratings, source-type classification | query, num_results,
site, date_restrict,
language, result_type |
1 credit |
| extract_webpage_content | One URL → clean markdown/text/HTML, boilerplate stripped, summary included | url, format,
full_content |
free |
| extract_multiple_webpages | Up to 5 URLs fetched and summarized concurrently | urls |
free |
| research_topic | Search → rank → dedupe → extract → structured synthesis prompt | topic, depth,
focus_areas |
1 credit per search (1 + one per focus area) |
All four tools are read-only: they retrieve and transform public web content and never modify anything.
3 · Example prompts
“Search for the latest rmcp release notes and summarize what changed in the newest version.”
Exercises google_search
with a follow-up extraction.
“Extract the content of https://modelcontextprotocol.io/specification and give me the key requirements for a compliant server.”
Exercises
extract_webpage_content.
“Research the current state of MCP server monetization — go to intermediate depth, focus on pricing models and payment providers, and cite your sources.”
Exercises
research_topic with focus areas.
4 · Credits & billing
One credit is one Google search; extraction is always free. New accounts start with a one-time 50-credit trial (30-day validity). The Connector plan ($15/month) includes 3,000 credits that reset each calendar month. Credit packs ($10 / $35 / $120) are one-time purchases valid for 6 months, drawn after any monthly allowance. Checkout and refunds are handled by Polar as merchant of record. Track usage and buy more in the portal.
5 · API keys
Each account holds exactly one active API key, issued automatically on your first portal visit and shown exactly once — we store only a hash, so it can never be re-displayed. If you lose it (or want to retire it), use Rotate key in the portal: the old key stops working within a minute and the replacement is shown once. Connector sign-ins from claude.ai/Desktop use OAuth and never need a key at all.
6 · Troubleshooting
“QUOTA_EXCEEDED” in a tool result
You've used your available credits. On the monthly plan, credits return at the calendar-month reset; trial credits are one-time and don't reset. Either way you can top up instantly in the portal — pack credits apply within seconds of checkout.
401 Unauthorized from Claude Code
The key in your header is wrong, rotated away, or revoked. Fetch your current key state in the portal and re-add the server with the fresh key.
My sign-in code never arrives
Codes come from admin@orchis.ai — check spam. Codes expire after 10 minutes and are limited to three per hour per address.