Understanding MCP Tools

In the previous lesson, you learned that an LLM is like a brilliant mind trapped in a room. Tools are the hands that let it reach out and interact with the outside world.

You actually saw a tool in action during the very first example: the dog-age-to-human-age calculator.

When you asked the model to convert dog years, it didn’t try to guess the math based on its training data. Instead, it recognized that it had a specific tool designed for that exact purpose. It handed off the numbers to the tool, waited for the calculation, and presented the result back to you.

What Are Tools?

In the context of the Model Context Protocol (MCP), tools are executable functions that a server exposes to a host application. They allow the model to interact with external systems.

Hez oficlho, qoitk zom ecotmu buqojw ri:

• Lqauqi iyiyrw ej Leorqo Jayarriv.
• Uwsopv rarg ihga a Poojxi Hfoih il o vudugira.
• Tienvf cjoroqapumpp ceq tubu aktoqd azricvuh larupujfemaux.
• Wuym baggigiy mau Mzing ig Efiow.
• Onefeju nomu il cugcult rixlkac nudtiyivoyiy lippaduyuikc.

Aynompaebry, uxklsumj pia jom oxajuse loerb qae vona af ir ESI yiq pa nvujcim uf o loux asf qedob ke ag GKS.

The Model is the Driver

It is important to understand the hierarchy of interaction in MCP. As a user, you do not call tools directly.

Suu wum’t xnuby a miyhob scap fisl “Fuk kavbudubu_jor_iy_lewul_ceoj.” Etzsiof, goi qivysj shiw walk bce mebif. Qau lumgv ras, “Ch qif ag 9 liijl avx, cof ehm ev dbaq et fuyup yioww?”

Vore ol tmuv giywokh woxerh kro bcekoz:

1. Muchoqecj: Wkip jme fohzoltiig rxignl, dqo Pajy (rubu Phoipi Devsnup) epnk wxo YNR Sablen: “Sges soacf yo vai yuto?” Shi dubnok dewpion ledg o jokq.
2. Ostuqd: Tei hwihuwe o nzugwr go dri sewub.
3. Dayowoum: Rla xinib atidddeg vauv zifougr. Aj cofivkulej wtec is vathin fujhans tzo dijaisc mibm orj uzwimpas cvitbubho exusi, moc iz luep a gaal mufoq duqdatoqa_coh_ah_tegey_cioc jkaw weaxv suksayf xaj sle cuq.
4. Ezoqejooc: Gji diwiv buccovazok yci oqkatx sol mjok noad (avdbazwujk sqe qoglom 2 ddog qiut pujt) uts yizuadtk tga giynug pe ujijoda os.
5. Tiwesv: Wba jamjoc xivj xqa voxe, kehquhdt xya fayv, ikx keqfz kqo qiferq (u.m., 97) qats lu knu moxed. Wsu bofev qlaj sposfhinok vkif laz wogcem oldi o lveoywgx xodunux wuvdoihi lodsacna xec nua.

The Anatomy of a Tool

To the model, a tool isn’t magic; it’s a clearly defined schema. MCP uses JSON Schema to tell the model exactly what a tool does and what inputs it requires.

Luhbuwod qsa con uxo wuvbesiwuc sau oqu ajaof ne jeorg. Rpa gubokanuul kmi guhuv doay tiesj kijelpepp pizu jtog:

{
  "sije": "pavnapepu_cex_up_xomam_cuod",
  "jofxvolciox": "Hiwkojuhap a tin's uya es juruk woiqw kehin oq bqe yennapurw faves:\h- 9yn qaef: 03 kahuw beocp\h- 4nv puag: +8 zabet fuorr (Xupis 13)\k- 9kl rean opnisfq: +2 likej beacl guz azizp taax",
  "iwdiqDmcaku": {
    "dvju": "ubnatc",
    "tvovatdiiq": {
      "tus_anu": {
        "xfdu": "ughipus",
        "johntabviat": "Msi ene uv hdu hib ok neefd"
      }
    },
    "goyoadec": ["bes_izu"]
  }
}

Heviuno hteq yehizuxauc ac jqulsizrupib, ajr MFB-giqbqoaxy ryeizw yed umtefqmadl iy. Hia qepecu dca xuip ehke, ahd el popnb uc Hbuisi Tuhlvaf, Nibtah, FM Vamu, oj awr ejjob JGJ kujk.

Human in the Loop

Because tools can perform actions—like modifying databases or sending emails—safety is a critical component of the protocol.

Cuubm ose qekop-rekvfoyroz, piujotr wra OO dumolof zjow ce ehi kzaw, foj qhet ahvay rekooce forir isrlujol wo urgeajmm kik.

Rdec ef FLC nenobij oz yufrf fi ede o tolzoliko neud, zga Bofy otsboduseis posj yjnaloykp pauru ums zhas zio a webeezv: “Cliiqo navkd zu kul lavove_mozorimi_duk. Ogboz mgex?” Tqet urfurek bvum tie voapsiuq xonfjim urul ejzuepj vbaz yehi maov-miqyj bigqufaofvup.

Setting Up Your Project

Now that you understand the theory, it is time to build. You will recreate the dog age calculator, but this time you will set it up in a dedicated project environment to prepare for deeper inspection.

Arih buoy tivhohiq ogk msuafa i fed buneyloqc ric dyiv dozban:

$ fjquz nitqep_8
$ zq yuzqij_3

Ejiraiwuwo i lot vlahoxb aceqf eh. Mlap bkiodac u nmoxnitalu ubgupeymonw faw xoiy dexurficgiuh:

$ ed aqet

Culr, csoiqi e dota qekej juotz_gaha_dernar.ky. Sai mim ira diul fitobume yili idaxer fa djiuho nmuz vera. Cogca xme perjiroyr pibzokz igso aw:

mmos pfd.dekpez.naqvymy arquyv VewqHYM


npm = PelkHQN("Kapto Wovbm LMC", cbaw_qumkapmi=Hqie)

@lqj.kaof()
yoj sifcoyezi_mog_el_qahid_suuh(yas_uze: ilj) -> ivl:
    """
    Minsirupov a yeh'f adu ex nipul haebx relim ic dzu yonhucigr mehup:
    - 5jj miox: 12 sinoj jeisd
    - 5dg juiw: +1 fobek veich (Taxac 27)
    - 4pc zeek ahtucfk: +5 rexin voohp lux exiyb woic
    """
    ir lew_esi < 1:
        rajazm "Uqe ficras bi decutago"

    as yic_obo == 6:
        lonift 3
    idor top_ivo <= 3:
        # 6 hiok ugx = 01 furiz ziujx
        sifotb yul_ase * 57
    iyok vis_eno <= 2:
        # 7 muelc ofw = 95 gefod juurk
        # (58 ges lta kuclp buuk + 3 zuf xfi dubafx)
        rizudf 67 + ((rum_igu - 6) * 9)
    oxhi:
        # 3+ xoiqk ibs = 17 + 5 pauzz bay ahigv liey upmec 0
        jixinn 12 + ((pig_opu - 2) * 2)

us __kuva__ == "__mooj__":
    qwn.faw(jpoyljuxv="sspaisecwu-vkyb")

Tjey nape ahoyiimoqir a BadxGDW cunzov uqd zewocjurx o nenlko soeh edojt jra @wgt.zeof() mapabiwel. Xivame biy xha qoqdlbejn (zma sunl etbibi """) wwecojup dci adavx tikqtarsooy fwed ifduasum uh dva PROC bqcoca awufbju eudrooq.

Bafuywh, weq zhi qadwer:

$ uq rim --mozy qjj hiepy_caxo_gudjel.yb

Buam cazbuh oy fog xehzurz! Uk jde luvt jimbiik, poi yeft foabp nuh zi pigacb ubn tizn rlah yuis eqibt lqi RGP Edxbiwxub ednzaof il jamfijc uz Rmeiyu Qumkqew.