An MCP server that hands coding agents structured, source-cited plant data for garden planning — species care profiles, mechanism-backed companion verdicts, and frost-date planting windows.
search_plants — look up species by common or scientific name, filtered by life cycle, edibility, or indoor suitability; returns candidate ids for use in the other tools.
plant_details — full care profile for one species: sun, water, hardiness zones, mature height, edibility, toxicity to people and pets, and known pests; withheld fields come back null and named in notes.
companion_check — decide whether two plants should share a bed, returning every mechanism behind the verdict: shared botanical family or overlapping pests (bad), documented Permapeople listing (good), with overridden reasons still shown.
planting_window — derive last spring frost, first autumn frost, season length, and USDA zone from ten years of observed daily minima at a location (coordinates or a zone), then check the plant's hardiness fit; add days_to_maturity and frost_tolerance to get sow-by and transplant dates.
Pass ids rather than names to avoid extra upstream requests; every tool returns compact JSON, no HTML or page dumps.
All tools are read-only and answer honestly — explaining gaps and refusing to guess spacing or crop timing rather than filling in folklore.
An experimental identify_plant (image URLs) exists but is off by default and not registered here.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@plant-intel-mcp is it safe to plant tomatoes next to potatoes? also check dill"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
An MCP server that gives coding agents structured plant data for garden planning: species care profiles, companion checks that explain themselves, and frost-date planting windows derived from ten years of observed weather.
Live demo: rootwise — an AI gardening advisor built on this server, showing every tool call and source it uses (source).
companion_check("tomato", "potato")
-> bad: both are Solanaceae. Same-family crops share soilborne diseases and
compete for the same rotation slot, so plant them apart and rotate the bed.A model asked about garden planning will happily produce plausible spacing numbers and companion-planting advice. Most of it is folklore, and none of it is traceable. Three things make this worth a server rather than a prompt:
The data is real and cited. Every record says where it came from and under what licence.
The gaps are visible. When an upstream tier withholds a field, the response says so. When no source publishes a number, it stays null with an explanation instead of being filled in.
Verdicts carry mechanisms. companion_check never says "these are good
together" because a gardening blog said so. It says why, from evidence in
the care data, and admits when it has no evidence at all.
Related MCP server: mcp-hazards
Tool | What it does |
| Search the species catalogue by name; returns candidates with ids. |
| Care profile: sun, water, hardiness zones, mature height, edibility, toxicity to people and pets, known pests. |
| Whether two plants belong near each other, with every mechanism behind the verdict. |
| Frost envelope, season length, and USDA zone fit for a site, from observed daily minima. |
| Experimental, off by default. Identify a plant from image URLs. Requires Perenual beta access. |
Every tool returns compact JSON. No HTML, no page dumps.
companion_check returns mechanisms, not folkloreNo API in this server's source roster publishes plant antagonists — Permapeople lists "companion to" links only — so a good/bad/neutral answer cannot simply be looked up. Rather than restate garden lore, the negative case is derived from evidence already in the care data:
Mechanism | Verdict | Reasoning |
| bad | Same botanical family: shared soilborne disease, same rotation slot. |
| bad | Overlapping pest susceptibility: co-planting concentrates the pest. |
| good | Documented in Permapeople (requires that source to be configured). |
A risk mechanism outranks a positive listing, but every matched reason is
returned, including the overridden one, so a mixed pair stays legible. A
neutral verdict states plainly that it is an absence of evidence, not
evidence of compatibility.
planting_window will not guess crop timingGive it coordinates and it derives, from ten whole years of Open-Meteo daily minima, the median last spring frost, first autumn frost, season length, and the site's USDA zone — then checks that against the plant's published hardiness range. Verified against published normals:
Site | Last frost | First frost | Season | Derived zone |
Minneapolis, MN | 04-28 | 10-28 | 183 days | 4b |
Portland, OR | 03-17 | 11-25 | 252 days | 8b |
Honolulu, HI | — | — | — | 13a (frost free) |
Sydney, AU | — | — | — | 10b (frost free) |
What it will not do is invent the crop half. No source in this roster
publishes days-to-maturity or a frost-hardiness class, so transplant and sow-by
dates appear only when you supply days_to_maturity and frost_tolerance
yourself. That division is deliberate: this server owns the climate half, and
the caller owns the crop half.
Southern-hemisphere sites are read on their own growing season. Grouping by calendar year splits a southern summer in two and reports a frost-free window of roughly one day.
Requires Node 20 or newer and a free Perenual API key.
npm install -g plant-intel-mcpOr run it without installing: npx plant-intel-mcp.
Or run it straight from the repo:
git clone https://github.com/wyattlindsey/plant-intel-mcp.git && cd plant-intel-mcp && npm install && npm run buildclaude mcp add plant-intel --env PERENUAL_API_KEY=your-key -- npx -y plant-intel-mcpIn claude_desktop_config.json:
{
"mcpServers": {
"plant-intel": {
"command": "npx",
"args": ["-y", "plant-intel-mcp"],
"env": {
"PERENUAL_API_KEY": "your-key"
}
}
}
}Variable | Required | Purpose |
| yes | Species data. Free key. |
| no | Adds documented companion listings. Free, self-service. |
| no | Set to |
| no | Cache location. Defaults under |
| no | Set to |
Missing the Perenual key does not stop the server. The tools stay listed and each call answers with the variable to set and where to get a key.
Stated plainly, because these constraints shape what the server can honestly answer.
Source | Licence | Limits that matter |
API terms; free tier is non-commercial | 100 requests/day. Free keys cover species ids 1–3000 only, not the advertised 10,000+ — and in practice most common garden crops sit above that line (tomato is 8759). Search covers the full catalogue; details do not, and an out-of-range request comes back as a 429 that this server translates rather than reporting as throttling. Care guides and hardiness maps are paid. Withheld fields return upgrade prompts, which this server strips. | |
CC BY-SA 4.0 | Optional, off unless configured. Positive companion links only — no antagonists. Its upstream includes PFAF and Kew, whose terms are more restrictive than CC BY-SA implies; check before relying on it commercially. | |
CC BY 4.0 | Free, no key. ERA5 reanalysis on a ~9 km grid — not station data. Expect frost dates to differ from local normals, especially in hills, valleys, or near water. Every |
Because of the 100/day ceiling, the server caches aggressively (30 days for species data, 90 for weather archives) and guards a daily budget, refusing to spend a request it does not have and naming the reset time when it declines. Cache hits are never charged against it.
plant_details never returns spacing. No source here publishes it, and a null
with a note is more useful than a number nobody stands behind.
npm test # unit and protocol tests, no network
npm run typecheck
npm run build
npm run test:live # opt-in; hits real APIs and spends quotaThe live suite exists for drift detection. Fixtures were built from published documentation, so it asserts that every field the mapper reads is still present upstream -- last confirmed against the live API on 2026-09-08. Its Open-Meteo tests need no credential and run anywhere.
See CLAUDE.md for architecture and conventions.
MIT. See LICENSE.
Data retrieved through this server remains under its own source's terms — see the table above. In particular, Perenual's free tier is non-commercial, and Permapeople's data is share-alike.
Whether two plants should share a bed, with the mechanism behind the answer. A "bad" verdict is derived from shared botanical family (rotation and soilborne disease) or overlapping pest susceptibility, never from folklore. A "good" verdict comes from a documented Permapeople listing when that source is configured. Every matched reason is returned, including any the verdict overrode.
Inputs: plant_a (required), plant_b (required).
Full care profile for one species: sun, water, hardiness zones, mature height, edibility, toxicity to people and pets, and known pests. Fields the configured Perenual plan withholds are reported as null and named in notes rather than left silently absent.
Inputs: plant (required).
Frost envelope and hardiness fit for a plant at a location. Given coordinates it derives last spring frost, first autumn frost, season length, and the USDA zone from ten years of observed daily minima. Supply days_to_maturity and frost_tolerance from your own crop data to also get transplant and sow-by dates -- no source this server reads publishes those, so they are never guessed.
Inputs: plant (required), latitude, longitude, percentile, +3 more.
Search the Perenual species catalogue by name and return candidate species with ids. Use the returned id with plant_details, companion_check, or planting_window rather than passing a name again -- resolving a name costs an extra upstream request.
Inputs: query (required), page, cycle, edible, +1 more.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Scored across 4 tools
Each tool targets a distinct operation: searching plants, retrieving details, computing planting windows, and checking companions. There is no overlap in purpose; the tools chain together via the search_then_detail flow, making selection unambiguous.
All tool names follow a consistent pattern: a noun ('plant') modified by a descriptor ('details', 'intel', 'window', 'check'). The names are clear, descriptive, and follow a predictable schema, with no mixed conventions or vague verbs.
Four tools is a lean but coherent set for a plant-intel domain. Each tool covers a distinct need (search, details, planting time, companionship), so none feel redundant. A fifth tool for bulk or batch queries might be additive, but the current count is appropriate for the niche scope.
The server covers the primary lifecycle: identify a plant, retrieve its profile, plan its planting window, and assess its compatibility with neighbors. Notable gaps include a missing tool for updating or managing the data (e.g., adding a custom plant or saving a garden plan), but the core research-oriented use case is fully covered.
Hosted MCP server for live public-data APIs and Skills for AI agents.
Agent-native MCP server over 49M+ US public and government records, privacy-first, always current.
Capability registry for the agentic economy. Semantic search over verified MCP server listings.
MCP server for generating rough-draft project plans from natural-language prompts.
