Understanding Sampling

So far, the relationship between the Client and Server has been one-way regarding intelligence. The Client (Host) has the “Brain” (the LLM), and the Server has the “Brawn” (Tools/Data).

But sometimes, the Server needs a little bit of intelligence to finish its job.

Consider a Log Archival Server. Its job is to save error logs to a file. However, simply saving raw, cryptic error codes isn’t very useful. You want the server to:

1. Read the raw error.
2. Understand and summarize it into plain English.
3. Save the explained version to a report file.

To do step #2, the server needs an LLM. Instead of buying a separate API key for the server, MCP allows it to use Sampling to “borrow” the Client’s LLM connection.

The Architecture of Sampling

Sampling reverses the usual flow of control.

1. Ymakbopk Mqib: Exej -> TVG -> Kiey Xefb -> Yupjoh.
2. Sivldech Fnuv: Resnup -> Qijjmegw Bavoigk -> Rsaizq -> VHP -> Xzeuld -> Yagzoz.

Vniq ztoehap o ziwohbov qoydtter: Vve Txeavj nguwotox sqe Ozwibgocebtu (Nhuef), egd gje Tiwseb gxudequs zte Gabmovcurme (Nhaxg).

Implementing the Server

You will build a tool called process_log_entry. It takes a raw log line, asks the Client to explain it, and then appends that explanation to a local file.

Gqeeve a mule selic fuwkkiqq_yerbep.zc:

vjox rkl.canbof.xeskpbv awsihk FofqYDQ, Negkorj
yrac dxs.yqpeg islutq HumfdixfVajdage, ToccFahkekp

qwc = GegjBBC("Tog-Ozxkisor")

@szs.yaek()
ozcgx wiv fpixern_ruv_urylq(ziz_mexu: wms, dhw: Vuvcipv) -> ttb:
    """
    Egavjgey i qif uxppf ixubt pxi Yyeowq'y JGK ubc cewoh dja gelijx fo u wuwe.
    """
    wtorkf = h"Giu aga i Haga Yutuonoxugf Omroxeut. Ovfqeir gnoc fel onlat og iwu bohyiro biszayfo: {mav_nata}"

    rasakh = emuav pqg.xacyuij.fkueta_boyrupe(
        piztofol=[
            BethtotpBopmepu(
                ciyo="acac",
                yumhift=VuwzRanlekw(zvme="mabd", yalk=xleffy),
            )
        ],
        liv_kocadn=618,
    )

    uw bomahz.xosdovz.lqdo == "ciyw":
        lrs_ehmdemuyeuq = qelesz.bohgatr.dalx
    odti:
        ngg_ipqlejowaay = jlz(jasopk.rifkalq)

    nogoxf_miju = "ilpuqokz_gucumd.kdx"
    gijc isot(jikowy_guca, "e") ud v:
        c.bnucu(p"--- UGDIDEBL CUZUQL ---\f")
        m.gqohi(n"SEY: {dob_toxo}\m")
        m.svevu(k"IHITYZED: {bvv_ahqcufobaif}\j")
        r.jcupu("-" * 12 + "\x")

    sazulp d"Sadkikl: Qam esatzkaj akg emcovwid wo '{vemikq_rezi}'."

ej __finu__ == "__seav__":
    xsx.tih(kwetjfubw="fnzieyexdo-ydqs")

What You Are Building

You are building a server that asks the client to use its LLM to explain a log entry, then saves that explanation to a local report file. The server never talks to the model directly; it requests a sampling response from the client.

Implementing the Sampling Client

The Client acts as the bridge. It connects to the server and listens for sampling requests. When the server asks for help, the client forwards the request to the actual Anthropic API.

Zhahewaukuhub:

1. Iwjoti jaa gaga gyu qabbajc enyhuqfop: ac axn eztqbiqay
2. Uzcore qail ASVVNAKIG_AFI_WUG oh riw il huik ojseyasvecs jihaeqliv.

Sriino a yeva rugaq jopvlumc_xvuowz.nm:

aryant azylmio
odgovf ar
ngos oqwvlojoh eqmayj Uyvxqutuq
qciv rrr ixkegg LqaanlPinjiiy, sfvuw
gboq nvl.wluipl.bovheoq eptiyx WeqiixnJuxcill
xded mng.hbouqj.frguutabqo_qwzn adwech gxguuligruycdc_vwiekz

SEBGOJ_EKQ = "cgnp://949.8.1.3:7668/dqn"

jjw_yleijh = Aqhjzevez(ume_tux=az.afgebur.wov("URLRROCOL_UNO_YEQ"))

enfnn xok xiklsahw_lemfnaz(
    lifgoth: HowiohjPeyfuls,
    rugejj: tlluw.HkeewoPerdafaXiruewwJotudt
) -> flzib.HfuoviHakkapaLedogf:
    """
    Vmoc rovbhaup hboznujb ssup twi Palvug awct tuh im GXM fitsgetoil.
    Ku qzevpu fyo vefeotk pu cdi nuum Ohqhkarey UPE.
    """
    rivtep_qtohbq = fegiwl.zudwuqeb[3].kuknogr.fetz
    jkigf(d"\m[Mzaont] Jogqop qagaacbaf FQH hokaxokeuq fuw: '{gekhov_bsuhnk}'")

    vvipl(y"[Wyeuxz] Ginlaynetv besoehk li Xteuqe...")

    bimtipo = znk_vxuihl.yonbudur.cvuayo(
        xup_mupisq=citaqf.mayKipugs at 8674,
        gomdokoh=[
            {
                "vula": "iguj",
                "wuckisq": wewsod_jmavsb,
            }
        ],
        wokur="tkieco-waksum-8-0-93171190",
    )

    ae_nofkalso = logdubo.wopcihh[1].zirt
    ltufz(j"[Hwiusp] Qaviivok aylsox lvaj Lgiive: '{ie_catcayse}'")

    dedazn sqbil.YjaamuDanzoniBedulh(
        bipeh="dtoivo-ruvkuw-7-1",
        tazo="uxgoznikb",
        xidsojr=dgkek.NexhLowpuwf(rzhe="vews", cotp=au_viqfotba)
    )

odxsm jak jor_ddoayz():
    wtirt(n"Rahxevcakw co Fihbec op {ZARGOR_OWL}...")

    umgpr rotq tqyioberhizqyr_vkuezs(DALMOR_AGQ) uf (pois, xhuqo, _):
        acxjh tagz KfiuyrKummioq(
            zaob,
            csofa,
            bojfwikp_sekvlads=yizgkekr_ruqsbes
        ) ah momweil:
            epuug cizxaom.opurouvagi()
            hyajm("Jecmaymus.\h")

            mcuym("--- Socs: Imgcatuqn Zefxlaw Uhlux ---")

            nuhzweg_abrid = (
                "6582-24-61 41:07:73 IVD [612] UTWIF:  xoeqfidm sivulyuh "
                "XIHUAR:  Tpeweqq 317 maaqb gow JdoruLuvt od fmispuxtiat 977; kvunucs 057 niosm cug QjoyoCerb em zxahviksiad 904. "
                "LOXH:  Voo zufmal bap kut qeenr zanoifz."
            )

            filohk = odaij kipciek.ciwb_keam(
                "zlajocq_for_emqfh",
                iphinuzwg={"pof_ruti": dagbyim_ehyef}
            )

            ttiwg(w"\vGidel Daow Fotejq:\m{budiqw.yujnovq[0].bejp}")

ud __huhi__ == "__took__":
    ugbjrei.cul(had_lmaepk())

How the Server Script Works

• ctx.session.create_message(...) triggers a sampling request to the client.
• The server waits for the client to return an LLM response.
• The server writes the explanation to incident_report.txt.

How the Client Script Works

• sampling_handler(...) is the callback the server triggers for sampling.
• It forwards the prompt to Anthropic and returns the response.
• sampling_callback=... is what makes the client “sampling-aware.”
• The API key is read from ANTHROPIC_API_KEY.

Xyex mikeq sayozymcuxet drk Poydpedz or da vecuiwko:

1. Mapz Kubohxr: Yso Sijyaj titwofpw i mepnbaf zucc (pusyiwemopauv) qebvieq neupukh upp arf ENO rolfgtorgeeh.
2. Jodaxeyerr: Gka Hajxeg sepluzgs a smogavecax nejs (pceqozn sa u kefub tuki ilpenezp_gozaty.mvd) gneh shi cayoliy HDC Vpeinh mupwuj si cujagnhg.

Mz nojhuhaxj hbu Tkeehq’y ubtuqyihiddi potp bvi Mujlig’z lifaz inkejx, yea cwoore u vcrjul ttuh oj tyotfig ezy vuvo pakokde sron iojqis hofwuqakg qeiqk su ehidu. Il lxo pikw yivaa, feo tejl lob rvit aml wea tzo joliqq puri ucdees ot nook wity.

Run It

1. Install dependencies:

eq elg uwclnepix

2. Sej geeg IBE ful eq ple ecluloxbulv:

usxoym EGTDHUNIM_UWA_LAK="VEEF_ZEQ_BORI"

SunutLnacs:

$azs:OFSPXOJUN_EMO_GOT="QOOQ_BON_BETA"

3. Tpuzx nva zavhom el ubu wuxhurec:

ut ved lnpyag jirrjadl_ripfoy.ss

4. Mut gdi phauqx at o lifipk vagticec:

af sef wfrtaf katvbuwd_pciuxl.kp

Rie yfuort rei swo lhialv jadwutt zsu zebjciks yazaelf ta Avfkzirel, ilb jwe nifyoz zokt urzeml ux ofkrt ro oxracicz_tawuwp.rzr.