@pipeworx/mcp-meps 0.1.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Mojibake Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,180 @@
+# meps
+
+What Americans actually take, and what for — US prescription drug use by medical
+condition, from AHRQ's Medical Expenditure Panel Survey.
+
+Part of [Pipeworx](https://pipeworx.io) — an MCP gateway connecting AI agents to 1573+ live data sources.
+
+MEPS surveys the whole US civilian non-institutionalized population: every payer,
+every age, commercially insured and uninsured alike. It is the only source in
+this catalog that links a prescription to the **condition it was written for**,
+which is what makes "what do people take for depression" answerable at all.
+
+## Tools
+
+| Tool | Answers |
+|---|---|
+| `meps_drugs_for_condition` | Which drugs Americans take for a condition, ranked by people treated |
+| `meps_conditions_for_drug` | What a drug is actually prescribed for — the off-label question |
+| `meps_drug_use` | One drug's national volume, total spend and out-of-pocket share |
+| `meps_top_drugs` | The most-used prescription drugs in the US for a year |
+
+Conditions are taken in plain words — `"depression"`, `"high cholesterol"`,
+`"type 2 diabetes"` — or as a CCSR code such as `MBD002`. Drug names are MEPS's
+cleaned **generic** names, so `atorvastatin`, not `Lipitor`.
+
+## These are survey estimates, and the sample is always in the response
+
+Every respondent carries a weight projecting them onto roughly 18,000 Americans.
+A cell backed by three people therefore produces a confident-looking
+"55,000 people" that means almost nothing. So:
+
+- every row carries **`sample_persons`** — the real respondents behind it;
+- `reliable: false` marks anything under **60** respondents, which is AHRQ's own
+  publication floor;
+- rows default to a minimum of **60** respondents — the same floor — so the
+  default answer contains nothing AHRQ would decline to publish. Lower it with
+  `min_sample` to see less-common drugs; those come back `reliable: false`.
+
+Treat the numbers as *rough magnitude and ranking*, not as counts. Ranking the
+top few drugs for a common condition is what this data is good for. Precise
+totals, small subgroups and year-over-year differences of a few percent are not.
+
+## Two things that would otherwise mislead
+
+**Fewer than half of prescriptions carry a condition.** 47.5% of fills in 2024
+are linked to any diagnosis; the rest are not attributed to one. So a condition
+total is a subset of prescribing, not a census of it, and an empty result can
+mean "not captured" rather than "not prescribed". Every response states the
+year's actual coverage rather than implying completeness.
+
+**One fill can be linked to several conditions.** That is correct for "is this
+drug used for this condition" and wrong for money: the condition tables carry no
+expenditure at all, because summing spend across conditions would invent
+national spending that was never spent. Money lives only in `meps_drug_use` and
+`meps_top_drugs`, where each fill is counted once.
+
+**Some "drugs" are therapeutic classes.** MEPS substitutes a class name for the
+product where naming it would identify a respondent — `INTERLEUKIN INHIBITORS`,
+`ANTINEOPLASTICS`, `ANTIDIABETIC AGENTS`. In 2024 that is 49 of 554 names, but
+those 49 carry **31.9% of all recorded spending**, so a ranking by cost is
+dominated by classes rather than products. `meps_drug_use` and `meps_top_drugs`
+say so in every response. A plural or category-sounding name is a class.
+
+## Data sources
+
+- MEPS public use files, three per year, from
+  <https://meps.ahrq.gov/data_files/pufs/> — Prescribed Medicines, the Appendix
+  to Event Files (the condition link), and Medical Conditions. Free, no
+  credential. The file numbers per year come from AHRQ's own index at
+  <https://meps.ahrq.gov/data_files/search_pufs.json>; they are not derivable
+  (2024 is HC-254A / HC-254I / HC-255), so they are looked up, never guessed.
+- Condition category labels from the HCUP CCSR reference file,
+  <https://hcup-us.ahrq.gov/toolssoftware/ccsr/dxccsr.jsp>.
+
+Years 2019 onward. Earlier files code conditions with CCS rather than CCSR and
+are not comparable.
+
+`.github/workflows/meps-refresh.yml` checks AHRQ's index on the 3rd of each month
+and loads any year that has all three files and is not already held. AHRQ
+publishes a year roughly annually without pre-announcing it, so the check does
+nothing 11 months out of 12 — that is cheaper than noticing a release late.
+
+**MEPStrends is not the source.** AHRQ's own pre-computed estimates at
+`meps.ahrq.gov/mepstrends/` sit behind HTTP Basic auth (`WWW-Authenticate: Basic
+realm="Secured Data Tools"`), so the estimates here are computed from the public
+files rather than read from theirs.
+
+## Quick Start
+
+Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):
+
+```json
+{
+  "mcpServers": {
+    "meps": {
+      "url": "https://gateway.pipeworx.io/meps/mcp"
+    }
+  }
+}
+```
+
+### What this endpoint actually serves
+
+`tools/list` at `https://gateway.pipeworx.io/meps/mcp` returns the tools in the table
+above **plus the shared Pipeworx meta-tools** — `ask_pipeworx`,
+`discover_tools`, `search_within`, `remember`/`recall` and the rest of the
+gateway-wide set. So the tool count you see is larger than this table: a
+single-pack endpoint currently lists roughly 30 shared tools alongside the
+pack's own. The connection's `initialize` response states its exact scope, and
+is the authoritative answer for a given day.
+
+This is deliberate, not multiplexing by accident. The meta-tools are what let a
+scoped connection answer a question this pack does not cover — via
+`ask_pipeworx`, which routes across the whole catalog — without you adding a
+second MCP server. There is currently no way to mount a pack endpoint without
+them; if the extra schemas cost you more context than the routing is worth,
+connect to the full gateway once rather than to several pack endpoints.
+
+Or connect to the full Pipeworx gateway to get every pack's tools listed
+directly, instead of just this one's:
+
+```json
+{
+  "mcpServers": {
+    "pipeworx": {
+      "url": "https://gateway.pipeworx.io/mcp"
+    }
+  }
+}
+```
+
+Both URLs reach the same gateway and the same 1573+ data sources. The
+only difference is which pack's tools are listed **directly**; `ask_pipeworx`
+reaches all of them from either one.
+
+## Standalone (no gateway account)
+
+This package also runs as a local stdio MCP server — no Pipeworx account, no
+gateway round-trip:
+
+```json
+{
+  "mcpServers": {
+    "meps": {
+      "command": "npx",
+      "args": ["-y", "@pipeworx/mcp-meps"]
+    }
+  }
+}
+```
+
+Or run it directly to confirm it starts:
+
+```bash
+npx -y @pipeworx/mcp-meps
+```
+
+It speaks MCP over stdin/stdout and answers `initialize`/`tools/list`/`tools/call`
+for **only** this pack's tools — none of the shared meta-tools the gateway
+connection above adds. Same source, same tools, no ask_pipeworx routing.
+
+## Using with ask_pipeworx
+
+Instead of calling tools directly, you can ask questions in plain English —
+this works on the pack endpoint above as well as on the full gateway:
+
+```
+ask_pipeworx({ question: "your question about Meps data" })
+```
+
+The gateway picks the right tool and fills the arguments automatically.
+
+## More
+
+- [Docs and guides](https://pipeworx.io/docs)
+- [pipeworx.io](https://pipeworx.io)
+
+## License
+
+MIT

package/bin/cli.js

@@ -0,0 +1,17 @@
+#!/usr/bin/env node
+//
+// Entry point for `npx @pipeworx/mcp-<slug>`.
+//
+// Packs ship as raw TypeScript (no build step — see publish-pack.sh for why:
+// tsx sidesteps every extensionless-import / bare-JSON-import edge case a
+// per-pack tsc build would have to solve one pack at a time). This file
+// registers tsx's ESM loader programmatically, then hands off to src/server.ts,
+// which wraps the pack's {tools, callTool} export in a stdio MCP server.
+//
+// Copied verbatim into every published pack repo by scripts/publish-pack.sh —
+// edit this file, not a per-pack copy.
+import { register } from 'tsx/esm/api';
+
+register();
+
+await import('../src/server.ts');

package/package.json

@@ -0,0 +1,27 @@
+{
+  "name": "@pipeworx/mcp-meps",
+  "version": "0.1.0",
+  "description": "MEPS — what Americans actually take, and what for.",
+  "type": "module",
+  "main": "src/index.ts",
+  "types": "src/index.ts",
+  "bin": {
+    "mcp-meps": "bin/cli.js"
+  },
+  "keywords": ["mcp", "mcp-server", "model-context-protocol", "pipeworx", "meps"],
+  "license": "MIT",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/pipeworx-io/mcp-meps.git"
+  },
+  "scripts": {
+    "typecheck": "tsc --noEmit"
+  },
+  "dependencies": {
+    "@modelcontextprotocol/sdk": "^1.30.0",
+    "tsx": "^4.19.0"
+  },
+  "devDependencies": {
+    "typescript": "^5.7.0"
+  }
+}

package/server.json

@@ -0,0 +1,18 @@
+{
+  "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json",
+  "name": "io.github.pipeworx-io/meps",
+  "title": "Meps",
+  "description": "MEPS — what Americans actually take, and what for.",
+  "version": "0.1.0",
+  "websiteUrl": "https://pipeworx.io/packs/meps",
+  "repository": {
+    "url": "https://github.com/pipeworx-io/mcp-meps",
+    "source": "github"
+  },
+  "remotes": [
+    {
+      "type": "streamable-http",
+      "url": "https://gateway.pipeworx.io/meps/mcp"
+    }
+  ]
+}

package/src/conditions.ts

@@ -0,0 +1,108 @@
+/**
+ * Turning what a caller says into a CCSR condition category.
+ *
+ * MEPS codes conditions as CCSR categories — END010, MBD002 — and nobody asks
+ * a question in those terms. An agent asks "what do people take for high
+ * cholesterol". Without this the pack answers zero rows to almost every real
+ * question, which reads as "we have no data" rather than "you used the wrong
+ * vocabulary".
+ *
+ * Two layers: these hand-written synonyms for the way people actually speak,
+ * then a trigram match against the official CCSR labels for everything else.
+ */
+
+/** Everyday phrasing -> CCSR categories. Several codes where the plain word
+ *  genuinely spans them: "diabetes" is END002 and END003, and returning only
+ *  one of them would quietly halve the answer. */
+export const CONDITION_SYNONYMS: Record<string, string[]> = {
+  // EVERY code below was checked against the HCUP CCSR reference file, because
+  // a first pass written from memory got roughly a dozen of them wrong in a way
+  // no test would catch: "epilepsy" pointed at cerebral palsy, "psoriasis" at
+  // pressure ulcer, "hypothyroidism" at diabetes. Each would have returned a
+  // confident, plausible, wrong list of drugs. If you add a synonym, look the
+  // code up — do not recall it.
+  'high cholesterol': ['END010'],
+  cholesterol: ['END010'],
+  hyperlipidemia: ['END010'],
+  'lipid disorder': ['END010'],
+  // END004/005/006 are diabetes categories too, but MEPS does not use them:
+  // in the 2024 file every diabetes person sits in END002 or END003. Including
+  // the unused codes would only widen the query, but naming them here would
+  // imply a coverage that is not in the data.
+  diabetes: ['END002', 'END003'],
+  'type 2 diabetes': ['END002', 'END003'],
+  'blood sugar': ['END002', 'END003'],
+  obesity: ['END009'],
+  'weight loss': ['END009'],
+  thyroid: ['END001'],
+  hypothyroidism: ['END001'],
+  'high blood pressure': ['CIR007'],
+  hypertension: ['CIR007'],
+  'heart failure': ['CIR019'],
+  'atrial fibrillation': ['CIR017'],
+  afib: ['CIR017'],
+  arrhythmia: ['CIR017'],
+  'coronary artery disease': ['CIR011'],
+  'pulmonary embolism': ['CIR013'],
+  depression: ['MBD002'],
+  depressive: ['MBD002'],
+  anxiety: ['MBD005'],
+  ocd: ['MBD006'],
+  ptsd: ['MBD007'],
+  bipolar: ['MBD003'],
+  schizophrenia: ['MBD001'],
+  psychosis: ['MBD001'],
+  'opioid use disorder': ['MBD018'],
+  'alcohol use disorder': ['MBD017'],
+  smoking: ['MBD024'],
+  'tobacco use': ['MBD024'],
+  insomnia: ['NVS016'],
+  'sleep disorder': ['NVS016'],
+  epilepsy: ['NVS009'],
+  seizures: ['NVS009'],
+  migraine: ['NVS010'],
+  headache: ['NVS010'],
+  'nerve pain': ['NVS019'],
+  asthma: ['RSP009'],
+  copd: ['RSP008'],
+  emphysema: ['RSP008'],
+  'acid reflux': ['DIG004'],
+  gerd: ['DIG004'],
+  heartburn: ['DIG004'],
+  arthritis: ['MUS006', 'MUS003'],
+  osteoarthritis: ['MUS006'],
+  'rheumatoid arthritis': ['MUS003'],
+  osteoporosis: ['MUS013'],
+  'back pain': ['MUS038'],
+  'low back pain': ['MUS038'],
+  'chronic pain': ['MUS010', 'MUS038', 'NVS019'],
+  'kidney disease': ['GEN003'],
+  ckd: ['GEN003'],
+  hiv: ['INF006'],
+  glaucoma: ['EYE003'],
+  // Deliberately absent: ADHD (CCSR splits it across neurodevelopmental and
+  // impulse-control in a way the data does not settle), eczema and psoriasis
+  // (no dedicated category — the nearest is "other specified inflammatory
+  // condition of skin"), and allergies (the only "allergic" category is an
+  // INJURY code for allergic reactions). Those fall through to the label
+  // search, which reports which category it matched so a caller can see what
+  // they actually got.
+};
+
+const CCSR_RE = /^[A-Z]{3}\d{3}$/;
+
+/** An exact CCSR code if the caller gave one, else the synonym hit, else null
+ *  meaning "fall through to a label search". */
+export function resolveConditionCodes(raw: string): string[] | null {
+  const t = raw.trim();
+  if (!t) return null;
+  if (CCSR_RE.test(t.toUpperCase())) return [t.toUpperCase()];
+  const lower = t.toLowerCase().replace(/\s+/g, ' ');
+  if (CONDITION_SYNONYMS[lower]) return CONDITION_SYNONYMS[lower];
+  // Longest containing synonym, so "type 2 diabetes mellitus" still resolves
+  // and does not lose to the shorter "diabetes".
+  const hit = Object.keys(CONDITION_SYNONYMS)
+    .filter((k) => lower.includes(k))
+    .sort((a, b) => b.length - a.length)[0];
+  return hit ? CONDITION_SYNONYMS[hit] : null;
+}