# faers.mobi — Claude Desktop / Cursor / Perplexity MCP

Paste-ready install. No API key. No pay gate.
Hosted (Streamable HTTP): https://faers.mobi/mcp
Stdio server: https://faers.mobi/mcp/server.py
Config: https://faers.mobi/mcp/claude_desktop.example.json
Human API: https://faers.mobi/API.md

FAERS disproportionality is a **hypothesis**, not causation.

Stdlib Python 3. Talks to live `GET https://faers.mobi/signals`. No new statistics.

## Hosted (preferred)

Add the URL. No local Python.

```json
{
  "mcpServers": {
    "faers-mobi": {
      "url": "https://faers.mobi/mcp"
    }
  }
}
```

Same tools as stdio. POST JSON-RPC (`initialize`, `tools/list`, `tools/call`).

`get_pipeline` asks https://globalpatientsafety.com/pipeline for one sponsor.
The rows are ClinicalTrials.gov registry facts, not a FAERS signal.
Omit status for ongoing studies. Omit role for lead sponsor only.
`role=any` includes collaborator-only trials.
Example: get_pipeline with sponsor Sanofi.

`get_vaers` asks https://vaers.globalpatientsafety.com/signals for a vaccine,
an event, or both. The rows are VAERS reporting signals, not a FAERS signal.
Counts are reporting, not incidence. Example: get_vaers with vaccine COVID19
and event Myocarditis.

`vaers_search_vaccine` lists pairs for one vaccine (optional event).
`vaers_get_signal` is one vaccine and one event, and returns the same `n`,
`eb05`, `eb05_peak_pre2021`, and `peak_in_covid_era` as GET /signals.
`vaers_get_profile` is format=profile for one vaccine. All four call that
same URL. `n_unique` is not in the VAERS extract. This server does not add
a second VAERS worker.

## Perplexity

Perplexity's web browsing finds the homepage but does not call `GET /signals`.
Add the hosted server as a custom connector instead:

1. Settings → Connectors → Add custom connector
2. URL: `https://faers.mobi/mcp`
3. Transport: Streamable HTTP
4. Auth: none
5. Enable the connector in a new chat

Example ask: "Retrieve the FAERS brief for tzield × Nausea. State the
data-through quarter. Hypothesis, not causation."

Custom connectors may need a paid Perplexity plan — check your plan.

## Local stdio

1. Save https://faers.mobi/mcp/server.py (stdlib only)
2. Merge https://faers.mobi/mcp/claude_desktop.example.json into Claude Desktop / Cursor MCP config
3. Set `args` to the absolute path of the file you saved
4. Leave `FAERS_MOBI_API` as `https://faers.mobi` (auth: none)

```json
{
  "mcpServers": {
    "faers-mobi": {
      "command": "python3",
      "args": ["/path/to/server.py"],
      "env": { "FAERS_MOBI_API": "https://faers.mobi" }
    }
  }
}
```

Smoke:

```
python3 server.py --call get_signal drug=tzield event=Nausea
python3 server.py --call rwe_brief drug=tzield event=Nausea
curl -sS -X POST https://faers.mobi/mcp \
  -H 'Content-Type: application/json' \
  -H 'Accept: application/json, text/event-stream' \
  -d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"get_signal","arguments":{"drug":"tzield","event":"Nausea"}}}'
```

## Tools (when to use)

- `search_drug` / `search_event` / `get_signal` — isolated pair lists (rows carry `indication` / `low_info`; pass `indication="hide"`, `low_info="hide"` for ranking asks)
- `get_signal_history` — quarterly n / EB05 / EWMA (`format=series`; needs drug AND event)
- `get_profile` — compact snapshot (`format=profile`; needs drug or event)
- `compare_class` — ATC4 peers (`format=class`; needs drug)
- `rwe_brief` — copyable markdown (`format=brief`; needs drug or event; drug+event adds class + series + explain). PDF is HTTP `format=pdf` (drug or event), not an MCP tool.
- `get_label_status` — cached FDA-label Novel status (`format=label`; needs drug AND event; no OpenFDA scrape)
- `explain_signal` / `find_novel_signals` — templates / filters; no new numbers
- `get_vaers` — VAERS vaccine × event pairs (`https://vaers.globalpatientsafety.com/signals`). Not a FAERS signal. `format` is json, series, profile, or class.
- `vaers_search_vaccine` — VAERS pairs for one vaccine. Optional event, min_eb05, min_n.
- `vaers_get_signal` — one VAERS vaccine and one event. Same n, eb05, eb05_peak_pre2021, and peak_in_covid_era as GET /signals. n_unique is not in the extract.
- `vaers_get_profile` — VAERS format=profile for one vaccine.

Always treat output as a statistical hypothesis from spontaneous reports, not evidence that a drug causes an event. Read n, eb05, and trend.

Ingredient grouping, product type and medication errors (2026-09-27)
- Every /signals row carries reported_drug (the raw FAERS string), ingredient
  (the active it resolves to), product_type (single | combination |
  trial_uncertain) and medication_error (true for use/supply PTs).
- product_type=single keeps only single-ingredient rows; medication_error=hide
  drops medication-error and product-issue rows. Both are opt-in VIEWS, never
  merges: unfiltered output is unchanged and every row carries its flags.
- Novelty is looked up on the ingredient, not the raw string.
- n and EB05 are per reported string and are NEVER summed or averaged across
  strings. Ingredient-level recomputation is a separate offline step.
- format=brief and format=pdf name every filter applied and the rows omitted
  by reason. format=class counts distinct ingredients, excluding the focal
  drug's own brands.