The Thom class and Thom isomorphism theorem for oriented vector bundles are proven ( at least to my knowledge) by induction on the open covers and some manipulation with Mayer-Vietoris sequences.
What is the "actual reason" behind the existence of Thom class? It seems strange that such an interesting class would exist just because some Mayer-Vietoris sequences routinely produce it.
It is easy to understand the existence of a Thom class by considering cellular cohomology. Let the given vector bundle be $E\to B$ with fibers of dimension $n$. One can assume without significant loss of generality that $B$ is a CW complex with a single 0-cell. The Thom space $T(E)$ is the quotient $D(E)/S(E)$ of the unit disk bundle of $E$ by the unit sphere bundle. One can give $T(E)$ a CW structure with $S(E)/S(E)$ as the only 0-cell and with an $(n+k)$-cell for each $k$-cell of $B$. These cells in $T(E)$ arise from pulling back the bundle $D(E)\to B$ via characteristic maps $D^k\to B$ for the $k$-cells of $B$. These pullback are products since $D^k$ is contractible.
In particular, $T(E)$ has a single $n$-cell and an $(n+1)$-cell for each 1-cell of $B$. There are no cells in $T(E)$ between dimension $0$ and $n$. The cellular boundary of an $(n+1)$-cell is 0 if $E$ is orientable over the corresponding 1-cell of $B$, and it is twice the $n$-cell in the opposite case. Thus $H^n(T(E);{\mathbb Z})$ is $\mathbb Z$ if $E$ is orientable and $0$ if $E$ is non-orientable. In the orientable case a generator of $H^n(T(E);{\mathbb Z})$ restricts to a generator of $H^n(S^n;{\mathbb Z})$ in the "fiber" $S^n$ of $T(E)$ over the 0-cell of $B$, hence the same is true for all the "fibers" $S^n$ and so one has a Thom class.
One not-very technical way to think of the Thom Isomorphism Theorem is that if you have a vector bundle, $p : E \to B$, if you remove the $0$-section $Z$ of the vector bundle from the Thom space $Th(p)$, you get a contractible space. So given a homology class in $H_* Th(p)$ the obstruction to trivializing it can be thought of as its intersection with $Z$. If there's no intersection, you're in the contractible space $Th(p) \setminus Z$. So the intersection of a homology class with $Z$ is tautologically the thing that keeps track of the homology class itself.
That's how I like to think of the Thom Isomorphism Theorem. So why is there a Thom class? Because you can intersect with $Z$. In cohomology this is cupping with the Thom class since that's what intersections translate to in cohomology.
You are thinking in terms of ordinary cohomology, where Mayer-Vietoris patches together the always present local orientation to produce a global one when you have it. It is more advanced, but maybe more illuminating, to note that the definition in general is intrinsically global. An $n$-plane bundle $p$ over a space $B$ has an associated sphere bundle $Sph(p)$ (by fiberwise one point compactification) with based fibers and thus a section. The quotient $Sph/B$ is the Thom space $T$ of $p$. For a multiplicative cohomology theory $E$, a Thom class $\mu$ is an element of $\tilde{E}^n(T)$ whose restriction to $\tilde{E}^n(S^n_b)\cong \tilde{E}^0(S^0)$ is a unit in this ring for any $b\in B$, where $S^n_b$ is the fiber over $b$ in $Sph(p)$. This definition is admitttedly mysterious. It suffices to give a Thom isomorphism and it is important geometrically, but the real explanation is more advanced and still not very well known. One should think of $E^*$ as represented by a ring spectrum $E$. Bundle theory naturally concerns spaces over $B$, or parametrized spaces. One can make sense of parametrized spectra over $B$, and one can even take the smash product of a parametrized space and a spectrum to obtain a parametrized spectrum. Thus one can make sense of $Sph(p)\wedge E$ as a spectrum over $B$. Of course, there is also a trivial spherical bundle $B\times S^n$ over $B$. It turns out that a Thom class as I defined it cohomologically is the same thing as a trivialization: an equivalence of parametrized spectra between $Sph(p)\wedge E$ and $(B\times S^n)\wedge E$. That is the geometric meaning. This is proven in the book Parametrized Homotopy Theory, by Sigurdsson and myself (available on my website).
There is a nice formulation and interpretation of the Thom isomorphism in terms of sheaf theory, or "Grothendieck's six functors". The statement of the Thom isomorphism in this setting is that if $f \colon E \to X$ is a rank $n$ orientable vector bundle, then $$ Rf_! \mathbf Z_E \cong \mathbf Z_X[-n].$$ In this form it is also trivial to prove. Note first that it's clear that $Rf_!\mathbf Z_E$ is concentrated in degree $n$, where it is a rank $1$ local system: the statement is local on $X$, so it is enough to prove this for a trivial bundle. So we should argue that this rank $1$ local system is trivial, i.e. that it has a global section. But a section of this local system is precisely an orientation of the vector bundle; again one argues locally, using that an orientation of the fiber $\mathbf R^n$ is the same thing as a generator of $H^n_c(\mathbf R^n,\mathbf Z)$.
So why does this global statement on $X$ imply the usual cohomological form of the Thom isomorphism? Let $j \colon E \to \overline E$ be the inclusion into the fiberwise one-point compactification, and $\overline f \colon \overline E \to X$ the projection. For any space $Y$ we let $a^Y$ be the projection from $Y$ to a point. The relative cohomology group $H^\bullet(\overline E, \overline E \setminus E)$ can be computed as $R a^{\overline E}_\ast j_! \mathbf Z_E$ (this is how you compute relative cohomology in general in the six functors language). But \begin{align*} Ra^{\overline E}_\ast j_! \mathbf Z_E & \simeq Ra^X_\ast R\overline f_\ast j_! \mathbf Z_E && \text{since $a^{\overline E} = a^X \circ \overline f$} \\ & \simeq Ra^X_\ast R\overline f_! j_!\mathbf Z_E && \text{since $\overline f$ is proper} \\ & \simeq Ra^X_\ast Rf_!\mathbf Z_E && \text{since $f = \overline f \circ j$} \\ & \simeq Ra^X_\ast \mathbf Z_X[-n] && \text{by the Thom isomorphism} \end{align*} and the final equation manifestly computes $H^{\bullet+n}(X)$.
The idea behind the Thom isomorphism $\beta:H^iX \rightarrow H^{n+i}(DE,SE)$ is implicit in the formula $$\int_{\sigma_{n+i}} \beta(\alpha_i) = \int_{X\cap \sigma_{n+i}} \alpha_i$$ Here $\sigma_{n+i}$ is a singular simplex in $DE$ and we have written integration for the evaluation of a cochain on a sum of simplices. Also $X\subset DE$ is identified with the zero-section.
The problem with this formula is that it doesn't make sense in full generality: after all $X\cap\sigma_{n+i}$ will not in general be a simplex again. And even if it is, it might be a simplex in many different ways (different parametrizations), so some choices must be made. These problems can be overcome and this is the "miracle" of the Thom isomorphism.
Note that the right hand side also requires an "orientation" of $X\cap\sigma_{n+i}$. This is why you also require an orientation on $E$.
For the Thom class $\tau = \beta(1)$ itself this gives the characterization $$\langle \tau, \sigma_n\rangle = \sharp ( X \cap \sigma_n )$$ where the intersection points are counted with appropriate signs. (In $DE$ a generic $n$-simplex has a zero-dimensional intersection with the zero section.)
You might find it helpful to learn something about Thom classes in other (generalized) cohomology theories: in de Rham cohomology and K-theory there are pretty explicit representatives for the respective Thom classes. And nothing beats the elegance of Thom classes in cobordism theories, where you've got a "tautological" Thom class.
Even the case of an oriented vector bundle over a point, which is where the story begins, is nontrivial. In this case the Thom isomorphism is the Poincare duality for the cohomology with compact supports on an oriented vector space. Ultimately, the Thom isomorphism theorem is a special form of the Poincare-Verdier duality. The fact that the Mayer-Vietoris technique is used in the proof is an indication that the Thom isomorphism deals with the cohomologies of some sheaves.
If the base of the vector bundle is compact and oriented, then the Thom isomorphism is equivalent to the Poincare-Lefschetz duality for an oriented manifold with boundary namely, the unit disk bundle determined by the vector bundle.
Yet another viewpoint on the Thom class.
Given an $n$-dimensional vector bundle $E\to B$ with connectd base $B$, there is fiber bundle pair $(D^n, S^{n-1}) \to (D(E), S(E))\to B$, where $D(E)$ is the unit disk bundle, and $S(E)$ is the unit sphere bundle. Then one may use the fibration long-exact sequences to show that $\pi_i(D^n,S^{n-1})\to \pi_i(D(E),S(E))$ is an isomorphism. Roughly, there is a long exact sequence $0=\pi_{i+1}(B,B)\to \pi_i(D^n,S^{n-1})\to \pi_i(D(E),S(E))\to \pi_{i}(B,B)=0$ (one can actually prove this using long exact sequences of each fibration and of the pairs). Hence by the relative Hurewicz theorem (see Theorem 4.37 of Hatcher), $\pi_i(D(E),S(E))=0$ for $i<n$, and hence $H_i(D(E),S(E))=0$ for $i<n$, and $H_n(D(E),S(E);\mathbb{Z})\cong \pi_n'(D(E),S(E))$, the quotient of $\pi_n(D(E),S(E)) \cong \mathbb{Z}$ by $\gamma x-x$, for $\gamma \in \pi_1(S(E))$, $x\in \pi_n(D(E),S(E))$, for $n\geq 2$. If $n=2$, $\pi_1(S(E))\twoheadrightarrow \pi_1(B)$, and $\pi_1(S(E))=\pi_1(B)$ for $n>2$, so this action will be non-trivial iff the vector bundle is non-orientable. Then $H_n(D(E),S(E))=\mathbb{Z}$ if $E$ is orientable, and $H_n(D(E),S(E))=\mathbb{Z}/2\mathbb{Z}$ if $E$ is non-orientable.
Now use the universal coefficient theorem to get the existence of the Thom class.
For $n=1$, $\pi_1(D^1, S^0)\to \pi_1(D(E),S(E))$ is still an isomorphism, but $\pi_1(D^1, S^0))$ is a 2-pointed set. We have a map $\pi_1(D(E),S(E)) \to H_1(D(E),S(E);\mathbb{Z}/2\mathbb{Z}$ which is an isomorphism (assuming the base $B$ is connected).
Once one has the Thom class, the Thom isomorphism maybe be proved as a consequence of the relative Leray-Hirsch theorem. The proof given in Appendix 4.D of Hatcher is by induction on skeleta; I don't know of a direct proof.
For $n=1$, $p:S(E)\to B$ is a 2-fold covering space, and the Gysin sequence may be deduced from the long exact sequence coming from the short exact sequence of chain complexes $$0 \to C*(B) \overset{p^*}{\to} C^*(S(E)) \overset{p^!}{\to} C^*(B)\to 0,$$ with coefficients in $\mathbb{Z}/2\mathbb{Z}$, where $p^!$ is the transfer map. The $(\mod 2)$-Euler class is the element in $H^1(B;\mathbb{Z}/2\mathbb{Z}) = Hom(\pi_1(B),\mathbb{Z}/2\mathbb{Z})$ defining the 2-fold cover $S(E)\to B$.
My intuition (maybe misleading?) of the Thom isomorphism is this: just as a vector bundle $\xi:X\to B$ is a twist of the trivial bundle, its Thom space $T\xi$ is a twist of the (rank $\xi$)-fold suspension of $B$ (Thom space of the trivial bundle is just the iterated suspension).
Now for a multiplicative cohomology theory $E$, it seems that this twist manifests itself in the fact that $\tilde E^*(T\xi)$ is a rank 1 projective module over $E^*(B)$, i. e. a twist of the free rank 1 module. And picking a $E^*$-orientation of $\xi$ is more or less the same as picking a generator (necessarily of degree rank $\xi$) of this module; in particular, such thing exists iff this module is free, and then the Thom isomorphism is clear - it is just dimension shift by degree of the generator.
Thus one may say that a bundle is $E$-orientable iff $E$ "is not confused by the twist of the iterated suspension introduced by the twist of the trivial bundle caused by $\xi$".
Thom class gives an orientation covector in every fiber $F\cong\mathbb R^n$ (of an oriented vector bundle) which can thought of a generator in $H^n(F-0)$ . Using local trivializations such covectors are defined locally. One needs to prove that these covectors glue to a cohomology class on the total space (with the zero section deleted), and this is where Mayer-Vietoris becomes relevant. How else would you glue? Read the exposition in Milnor-Stasheff or Bott-Tu.
One simple way of understanding the Thom class is just by "thickening" in homology, which is essentially what Hatcher wrote. Suppose that $B$ is arc-connected and $p: E \to B$ is an oriented vector bundle of rank $n$. If $Z$ denotes the zero section, the Thom space is $E$ rel $E-Z$, which is the same as the disc bundle of $E$ relative to its boundary. The usual definition of the fact that $E$ is oriented goes through the compatibility of the local charts for $E$, but it is the same as saying that, for any continuous path $\gamma$ in $B$, the monodromy sends the orientation of $(E_{\gamma_0}, E_{\gamma(0)} - Z_{\gamma(0)})$ to the orientation of $(E_{\gamma_1}, E_{\gamma(1)} - Z_{\gamma(1)})$. In this context, the Thom class $c$ in $H_n(E, E-Z)$ is just the fundamental class of $(E_p, E_p - Z_p)$ at any point $p \in B$, seen as embedded in $(E, E-Z)$. By the monodromy above, this choice is independent of $p$.
Finally, the Thom isomorphism $H_k(B) \to H_{n+k}(E, E-Z)$ is given by triangulating the inverse image by $p$ of any class $[z] \in H_k(B)$. Actually, this inverse image is a kind of product between $c$ and $z$. The interest here is that this product is well-defined even if the bundle is not trivial. This product is easier to define in cohomology, but in any case, we do not need this to define the Thom isomorphism : it is just thickening.
Of course, one needs to say what mean the words "by triangulating the inverse image..." above. Well, one can make this easily rigorous. Indeed, if $B$ is a smooth closed manifold and the coefficients ring is the real numbers R, Thom proved that any homology class in $H_*(B;R)$ contains a reprentative made of a finite linear combination of closed oriented submanifolds with real coefficients. Then the inverse image by $p$ of a smooth submanifold $M$ is a smooth submanifold of dimension $*+n$ of $E$. Then just see it in the unit disc bundle relative to the sphere bundle. This gives a homology class in the Thom space.
