# @pipeworx/amf-filings

French AMF-regulated company disclosures (résultats, franchissements de
seuils, information privilégiée, rachats d'actions, etc.) from Info-financière,
the French Officially Appointed Mechanism (OAM) — updated as the AMF publishes,
not months behind like the community ESEF index.

Part of [Pipeworx](https://pipeworx.io) — an MCP gateway connecting AI agents to 1861+ live data sources. This is an independent, unofficial integration — not affiliated with, endorsed by, or published by the upstream provider.

## Tools

- `amf_search_filings(company?, isin?, lei?, date_from?, date_to?, language?, limit?, offset?)`
  — search disclosures by issuer (ISIN preferred, LEI often null, or free-text
  name) and/or a publication date window. `date_to` is INCLUSIVE (the last
  day you want included) — set `date_from` and `date_to` to the SAME date
  for a single-day check. Returns the newest first.
- `amf_get_filing(id)` — fetch one disclosure by its Info-financière record id
  (`uin_idt_uin`), as returned by `amf_search_filings`.

## Why this pack exists

`esef-filings` proxies `filings.xbrl.org`, a community index whose current-year
coverage for French issuers is effectively absent — a named external developer
(evaluating us for a weekly equity-analysis pipeline) hit this directly and
rejected the pack over it. Measured live 2026-09-23, before this pack existed:

    France, ESEF index (filings.xbrl.org):  2023: ~ok   2024: ~ok   2025: near-zero

This pack goes to the primary source instead of the aggregator. **Measured
freshness, 2026-09-23**: the newest record in the dataset was timestamped
~1 minute before the query that found it. A `date_from`/`date_to` window of
"yesterday to today" reliably returns same-day filings — verified with a live
call against the deployed gateway (see `workers/gateway/src/tool-examples.json`
for the exact arguments used).

## One call per day, not one call per issuer

`company`, `isin`, and `lei` are all optional. Omit them entirely and pass
only `date_from`/`date_to` to get every issuer's filings for that window in
ONE call (paginate with `limit`/`offset` only if a single day's market-wide
volume exceeds the 100-row cap) — verified live 2026-10-19. A design partner
doing a daily check across ~155 issuers was instead calling once per issuer
and exhausting a 50-credit-per-call allowance in ~35 calls (fleet #3300); the
date-window call is the cheap way to do that check.

## ⚠️ Six things a caller must know

1. **This is a document index, not an XBRL-facts API.** Every row is a
   disclosure EVENT: a title, an issuer, a publication timestamp, and a link
   to the underlying PDF (occasionally an ESEF/XBRL package). There is no
   per-fact extraction — no "give me net income for FY2025" lookup. Whether
   `filings.xbrl.org`-shaped structured facts are reachable for these same
   issuers via this OAM was **not verified** while building this pack; treat
   it as a separate, open question, not a promise this pack makes.
2. **LEI is frequently null.** `identificationsociete_iso_cd_lei` is populated
   on many but not all rows. ISIN (`identificationsociete_iso_cd_isi`) is far
   more reliable — prefer it as the join key. Do not build an LEI-only
   pipeline against this pack; a miss on LEI does not mean "no filings."
3. **Not scoped to France-domiciled issuers.** The feed carries any issuer
   that discloses through a French disseminator — this includes non-French
   subsidiaries of French groups (e.g. Gabon-domiciled TotalEnergies EP
   Gabon). `country` reflects the issuer's own registered country in the
   AMF's record, not "France" by default.
4. **`company` matches the LEGAL name, not the trade name — with an
   automatic fallback.** The AMF stores issuers under their registered legal
   name (e.g. "TELEVISION FRANCAISE 1"), and `search()` is token-based, not
   substring, so a trade-name query like `company: "TF1"` matches zero rows
   there even though 38 genuinely exist in the same window (verified live
   2026-10-19, fleet #3300). `amf_search_filings` now retries against the
   disclosure TITLE when the legal-name search comes back empty, and the
   response carries `matched_via: "title"` plus `company_resolved` (the real
   legal name) when that is what found it — pass `isin` when you have it to
   skip the fallback entirely.
5. **`subtype_of_information` is normalised to a real `null`.** Some
   disseminators (307 in particular, 9,519 rows at last check) send the
   literal STRING `"NULL"` rather than an actual null. This pack converts any
   case-insensitive `"null"` string to a real null before returning it.
6. **`date_to` is INCLUSIVE of that whole UTC day.** A one-day check sets
   `date_from` and `date_to` to the SAME date; this pack handles the
   upstream's own exclusive bound internally (querying one calendar day
   past it) and echoes the actual bound sent as `upstream_date_to_exclusive`.
   This used to not be true: a same-day window
   (`date_from=date_to="2026-10-16"`) silently returned zero rows where 61
   existed, because the upstream bound was passed straight through as an
   exclusive one — fixed 2026-10-20 (a paying customer running one-day ESEF
   checks across every issuer hit it directly). A `date_from` after
   `date_to` now returns `empty_reason: "empty_window"` with an explanatory
   `message` instead of a bare zero.

Bilingual issuers file the same disclosure in French and English as two
separate rows (each with its own `id`) within the same minute —
`amf_search_filings` does not merge these; pass `language` to pick one
edition, or group results client-side on the returned `language_pair_key`
(company + type/subtype + publication minute).

## Auth

Keyless. No account, no API key.

## Data sources

- <https://www.info-financiere.gouv.fr/api/explore/v2.1/catalog/datasets/flux-amf-new-prod/records>
  — Opendatasoft Explore v2.1 records API, `flux-amf-new-prod` dataset
  (500,000+ records at time of writing). This is the primary AMF disclosure
  flux, not a mirror or cache of it — Pipeworx queries it live on every call.

## Gotchas

- The historic host `www.info-financiere.fr` 301-redirects every request to
  `www.info-financiere.gouv.fr` — this pack calls the `.gouv.fr` host
  directly to save the round trip.
- `limit` is capped at 100 by the upstream API (`-1 <= limit <= 100`); a
  higher value 400s rather than silently clamping. This pack clamps to 100
  before the request goes out and reports the original value as
  `limit_clamped_from` in the response when it does, so the clamp is never
  silent.
- ODSQL `where` clauses on `uin_dat_amf` (the filing timestamp) genuinely
  bind — verified live with a same-day-exclusion test — unlike the sibling
  `esef-filings` date-filter bug this pack was written to not repeat.
- `search()` on the company-name field is a loose token match (per ODSQL
  semantics), not an exact-name filter — a query for one issuer can surface a
  related entity with a similar name (e.g. a subsidiary). Prefer `isin` when
  you have it.

## Quick Start

Add to your MCP client (Claude Desktop, Cursor, Windsurf, etc.):

```json
{
  "mcpServers": {
    "amf-filings": {
      "url": "https://gateway.pipeworx.io/amf-filings/mcp"
    }
  }
}
```

### What this endpoint actually serves

`tools/list` at `https://gateway.pipeworx.io/amf-filings/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 1861+ data sources. The
only difference is which pack's tools are listed **directly**; `ask_pipeworx`
reaches all of them from either one.

## No MCP client? Call it over HTTP

```bash
curl -X POST https://gateway.pipeworx.io/v1/tools/amf_search_filings \
  -H 'Content-Type: application/json' \
  -d '{"isin":"FR0013230950","limit":3}'
```

No account needed for the first calls. Inspect any tool: `GET https://gateway.pipeworx.io/v1/tools/amf_search_filings`. Find one: `POST https://gateway.pipeworx.io/v1/tools/search_packs` with `{"query":"..."}`.

## Run it from npm

`@pipeworx/mcp-amf-filings` is a local stdio bridge to the Pipeworx gateway: your MCP client
starts it with `npx`, and it forwards every call to the same
`https://gateway.pipeworx.io/amf-filings/mcp` endpoint shown above. It needs network access;
there is no offline mode.

```json
{
  "mcpServers": {
    "amf-filings": {
      "command": "npx",
      "args": ["-y", "@pipeworx/mcp-amf-filings"]
    }
  }
}
```

Without a key it calls anonymously, capped at 50 calls a day. A free account
raises that to 200 a day: get a key at https://pipeworx.io/signup and add
`"env": { "PIPEWORX_API_KEY": "<your key>" }` to the server entry.
The key is sent as a Bearer token, the same way a hosted connection sends it,
so calls are metered and billed exactly as they are there.

It lists **only** this pack's tools by default. Set `PIPEWORX_META_TOOLS=1` to
also list `ask_pipeworx` and the other gateway-wide tools.

Behind an HTTPS proxy: Node's built-in `fetch` does not read `HTTPS_PROXY` by
default. Newer Node releases honour it when `NODE_USE_ENV_PROXY=1` is set; check
your Node version's documentation.

## 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 Amf Filings 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